1 /*
2 * Copyright (c) 1997, 2010, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "precompiled.hpp"
26 #include "interpreter/interpreter.hpp"
27 #include "interpreter/interpreterRuntime.hpp"
28 #include "interpreter/templateTable.hpp"
29 #include "memory/universe.inline.hpp"
30 #include "oops/methodDataOop.hpp"
31 #include "oops/objArrayKlass.hpp"
32 #include "oops/oop.inline.hpp"
33 #include "prims/methodHandles.hpp"
34 #include "runtime/sharedRuntime.hpp"
35 #include "runtime/stubRoutines.hpp"
36 #include "runtime/synchronizer.hpp"
37
38 #ifndef CC_INTERP
39 #define __ _masm->
40
41 //----------------------------------------------------------------------------------------------------
42 // Platform-dependent initialization
43
44 void TemplateTable::pd_initialize() {
45 // No i486 specific initialization
46 }
47
48 //----------------------------------------------------------------------------------------------------
49 // Address computation
50
51 // local variables
52 static inline Address iaddress(int n) {
53 return Address(rdi, Interpreter::local_offset_in_bytes(n));
54 }
55
56 static inline Address laddress(int n) { return iaddress(n + 1); }
57 static inline Address haddress(int n) { return iaddress(n + 0); }
58 static inline Address faddress(int n) { return iaddress(n); }
59 static inline Address daddress(int n) { return laddress(n); }
60 static inline Address aaddress(int n) { return iaddress(n); }
61
62 static inline Address iaddress(Register r) {
63 return Address(rdi, r, Interpreter::stackElementScale());
64 }
65 static inline Address laddress(Register r) {
66 return Address(rdi, r, Interpreter::stackElementScale(), Interpreter::local_offset_in_bytes(1));
67 }
68 static inline Address haddress(Register r) {
69 return Address(rdi, r, Interpreter::stackElementScale(), Interpreter::local_offset_in_bytes(0));
70 }
71
72 static inline Address faddress(Register r) { return iaddress(r); }
73 static inline Address daddress(Register r) { return laddress(r); }
74 static inline Address aaddress(Register r) { return iaddress(r); }
75
76 // expression stack
77 // (Note: Must not use symmetric equivalents at_rsp_m1/2 since they store
78 // data beyond the rsp which is potentially unsafe in an MT environment;
79 // an interrupt may overwrite that data.)
80 static inline Address at_rsp () {
81 return Address(rsp, 0);
82 }
83
84 // At top of Java expression stack which may be different than rsp(). It
85 // isn't for category 1 objects.
86 static inline Address at_tos () {
87 Address tos = Address(rsp, Interpreter::expr_offset_in_bytes(0));
88 return tos;
89 }
90
91 static inline Address at_tos_p1() {
92 return Address(rsp, Interpreter::expr_offset_in_bytes(1));
93 }
94
95 static inline Address at_tos_p2() {
96 return Address(rsp, Interpreter::expr_offset_in_bytes(2));
97 }
98
99 // Condition conversion
100 static Assembler::Condition j_not(TemplateTable::Condition cc) {
101 switch (cc) {
102 case TemplateTable::equal : return Assembler::notEqual;
103 case TemplateTable::not_equal : return Assembler::equal;
104 case TemplateTable::less : return Assembler::greaterEqual;
105 case TemplateTable::less_equal : return Assembler::greater;
106 case TemplateTable::greater : return Assembler::lessEqual;
107 case TemplateTable::greater_equal: return Assembler::less;
108 }
109 ShouldNotReachHere();
110 return Assembler::zero;
111 }
112
113
114 //----------------------------------------------------------------------------------------------------
115 // Miscelaneous helper routines
116
117 // Store an oop (or NULL) at the address described by obj.
118 // If val == noreg this means store a NULL
119
120 static void do_oop_store(InterpreterMacroAssembler* _masm,
121 Address obj,
122 Register val,
123 BarrierSet::Name barrier,
124 bool precise) {
125 assert(val == noreg || val == rax, "parameter is just for looks");
126 switch (barrier) {
127 #ifndef SERIALGC
128 case BarrierSet::G1SATBCT:
129 case BarrierSet::G1SATBCTLogging:
130 {
131 // flatten object address if needed
132 // We do it regardless of precise because we need the registers
133 if (obj.index() == noreg && obj.disp() == 0) {
134 if (obj.base() != rdx) {
135 __ movl(rdx, obj.base());
136 }
137 } else {
138 __ leal(rdx, obj);
139 }
140 __ get_thread(rcx);
141 __ save_bcp();
142 __ g1_write_barrier_pre(rdx, rcx, rsi, rbx, val != noreg);
143
144 // Do the actual store
145 // noreg means NULL
146 if (val == noreg) {
147 __ movptr(Address(rdx, 0), NULL_WORD);
148 // No post barrier for NULL
149 } else {
150 __ movl(Address(rdx, 0), val);
151 __ g1_write_barrier_post(rdx, rax, rcx, rbx, rsi);
152 }
153 __ restore_bcp();
154
155 }
156 break;
157 #endif // SERIALGC
158 case BarrierSet::CardTableModRef:
159 case BarrierSet::CardTableExtension:
160 {
161 if (val == noreg) {
162 __ movptr(obj, NULL_WORD);
163 } else {
164 __ movl(obj, val);
165 // flatten object address if needed
166 if (!precise || (obj.index() == noreg && obj.disp() == 0)) {
167 __ store_check(obj.base());
168 } else {
169 __ leal(rdx, obj);
170 __ store_check(rdx);
171 }
172 }
173 }
174 break;
175 case BarrierSet::ModRef:
176 case BarrierSet::Other:
177 if (val == noreg) {
178 __ movptr(obj, NULL_WORD);
179 } else {
180 __ movl(obj, val);
181 }
182 break;
183 default :
184 ShouldNotReachHere();
185
186 }
187 }
188
189 Address TemplateTable::at_bcp(int offset) {
190 assert(_desc->uses_bcp(), "inconsistent uses_bcp information");
191 return Address(rsi, offset);
192 }
193
194
195 void TemplateTable::patch_bytecode(Bytecodes::Code bytecode, Register bc,
196 Register scratch,
197 bool load_bc_into_scratch/*=true*/) {
198
199 if (!RewriteBytecodes) return;
200 // the pair bytecodes have already done the load.
201 if (load_bc_into_scratch) {
202 __ movl(bc, bytecode);
203 }
204 Label patch_done;
205 if (JvmtiExport::can_post_breakpoint()) {
206 Label fast_patch;
207 // if a breakpoint is present we can't rewrite the stream directly
208 __ movzbl(scratch, at_bcp(0));
209 __ cmpl(scratch, Bytecodes::_breakpoint);
210 __ jcc(Assembler::notEqual, fast_patch);
211 __ get_method(scratch);
212 // Let breakpoint table handling rewrite to quicker bytecode
213 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::set_original_bytecode_at), scratch, rsi, bc);
214 #ifndef ASSERT
215 __ jmpb(patch_done);
216 #else
217 __ jmp(patch_done);
218 #endif
219 __ bind(fast_patch);
220 }
221 #ifdef ASSERT
222 Label okay;
223 __ load_unsigned_byte(scratch, at_bcp(0));
224 __ cmpl(scratch, (int)Bytecodes::java_code(bytecode));
225 __ jccb(Assembler::equal, okay);
226 __ cmpl(scratch, bc);
227 __ jcc(Assembler::equal, okay);
228 __ stop("patching the wrong bytecode");
229 __ bind(okay);
230 #endif
231 // patch bytecode
232 __ movb(at_bcp(0), bc);
233 __ bind(patch_done);
234 }
235
236 //----------------------------------------------------------------------------------------------------
237 // Individual instructions
238
239 void TemplateTable::nop() {
240 transition(vtos, vtos);
241 // nothing to do
242 }
243
244 void TemplateTable::shouldnotreachhere() {
245 transition(vtos, vtos);
246 __ stop("shouldnotreachhere bytecode");
247 }
248
249
250
251 void TemplateTable::aconst_null() {
252 transition(vtos, atos);
253 __ xorptr(rax, rax);
254 }
255
256
257 void TemplateTable::iconst(int value) {
258 transition(vtos, itos);
259 if (value == 0) {
260 __ xorptr(rax, rax);
261 } else {
262 __ movptr(rax, value);
263 }
264 }
265
266
267 void TemplateTable::lconst(int value) {
268 transition(vtos, ltos);
269 if (value == 0) {
270 __ xorptr(rax, rax);
271 } else {
272 __ movptr(rax, value);
273 }
274 assert(value >= 0, "check this code");
275 __ xorptr(rdx, rdx);
276 }
277
278
279 void TemplateTable::fconst(int value) {
280 transition(vtos, ftos);
281 if (value == 0) { __ fldz();
282 } else if (value == 1) { __ fld1();
283 } else if (value == 2) { __ fld1(); __ fld1(); __ faddp(); // should do a better solution here
284 } else { ShouldNotReachHere();
285 }
286 }
287
288
289 void TemplateTable::dconst(int value) {
290 transition(vtos, dtos);
291 if (value == 0) { __ fldz();
292 } else if (value == 1) { __ fld1();
293 } else { ShouldNotReachHere();
294 }
295 }
296
297
298 void TemplateTable::bipush() {
299 transition(vtos, itos);
300 __ load_signed_byte(rax, at_bcp(1));
301 }
302
303
304 void TemplateTable::sipush() {
305 transition(vtos, itos);
306 __ load_unsigned_short(rax, at_bcp(1));
307 __ bswapl(rax);
308 __ sarl(rax, 16);
309 }
310
311 void TemplateTable::ldc(bool wide) {
312 transition(vtos, vtos);
313 Label call_ldc, notFloat, notClass, Done;
314
315 if (wide) {
316 __ get_unsigned_2_byte_index_at_bcp(rbx, 1);
317 } else {
318 __ load_unsigned_byte(rbx, at_bcp(1));
319 }
320 __ get_cpool_and_tags(rcx, rax);
321 const int base_offset = constantPoolOopDesc::header_size() * wordSize;
322 const int tags_offset = typeArrayOopDesc::header_size(T_BYTE) * wordSize;
323
324 // get type
325 __ xorptr(rdx, rdx);
326 __ movb(rdx, Address(rax, rbx, Address::times_1, tags_offset));
327
328 // unresolved string - get the resolved string
329 __ cmpl(rdx, JVM_CONSTANT_UnresolvedString);
330 __ jccb(Assembler::equal, call_ldc);
331
332 // unresolved class - get the resolved class
333 __ cmpl(rdx, JVM_CONSTANT_UnresolvedClass);
334 __ jccb(Assembler::equal, call_ldc);
335
336 // unresolved class in error (resolution failed) - call into runtime
337 // so that the same error from first resolution attempt is thrown.
338 __ cmpl(rdx, JVM_CONSTANT_UnresolvedClassInError);
339 __ jccb(Assembler::equal, call_ldc);
340
341 // resolved class - need to call vm to get java mirror of the class
342 __ cmpl(rdx, JVM_CONSTANT_Class);
343 __ jcc(Assembler::notEqual, notClass);
344
345 __ bind(call_ldc);
346 __ movl(rcx, wide);
347 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::ldc), rcx);
348 __ push(atos);
349 __ jmp(Done);
350
351 __ bind(notClass);
352 __ cmpl(rdx, JVM_CONSTANT_Float);
353 __ jccb(Assembler::notEqual, notFloat);
354 // ftos
355 __ fld_s( Address(rcx, rbx, Address::times_ptr, base_offset));
356 __ push(ftos);
357 __ jmp(Done);
358
359 __ bind(notFloat);
360 #ifdef ASSERT
361 { Label L;
362 __ cmpl(rdx, JVM_CONSTANT_Integer);
363 __ jcc(Assembler::equal, L);
364 __ cmpl(rdx, JVM_CONSTANT_String);
365 __ jcc(Assembler::equal, L);
366 __ stop("unexpected tag type in ldc");
367 __ bind(L);
368 }
369 #endif
370 Label isOop;
371 // atos and itos
372 // String is only oop type we will see here
373 __ cmpl(rdx, JVM_CONSTANT_String);
374 __ jccb(Assembler::equal, isOop);
375 __ movl(rax, Address(rcx, rbx, Address::times_ptr, base_offset));
376 __ push(itos);
377 __ jmp(Done);
378 __ bind(isOop);
379 __ movptr(rax, Address(rcx, rbx, Address::times_ptr, base_offset));
380 __ push(atos);
381
382 if (VerifyOops) {
383 __ verify_oop(rax);
384 }
385 __ bind(Done);
386 }
387
388 // Fast path for caching oop constants.
389 // %%% We should use this to handle Class and String constants also.
390 // %%% It will simplify the ldc/primitive path considerably.
391 void TemplateTable::fast_aldc(bool wide) {
392 transition(vtos, atos);
393
394 if (!EnableMethodHandles) {
395 // We should not encounter this bytecode if !EnableMethodHandles.
396 // The verifier will stop it. However, if we get past the verifier,
397 // this will stop the thread in a reasonable way, without crashing the JVM.
398 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
399 InterpreterRuntime::throw_IncompatibleClassChangeError));
400 // the call_VM checks for exception, so we should never return here.
401 __ should_not_reach_here();
402 return;
403 }
404
405 const Register cache = rcx;
406 const Register index = rdx;
407
408 resolve_cache_and_index(f1_oop, rax, cache, index, wide ? sizeof(u2) : sizeof(u1));
409 if (VerifyOops) {
410 __ verify_oop(rax);
411 }
412 }
413
414 void TemplateTable::ldc2_w() {
415 transition(vtos, vtos);
416 Label Long, Done;
417 __ get_unsigned_2_byte_index_at_bcp(rbx, 1);
418
419 __ get_cpool_and_tags(rcx, rax);
420 const int base_offset = constantPoolOopDesc::header_size() * wordSize;
421 const int tags_offset = typeArrayOopDesc::header_size(T_BYTE) * wordSize;
422
423 // get type
424 __ cmpb(Address(rax, rbx, Address::times_1, tags_offset), JVM_CONSTANT_Double);
425 __ jccb(Assembler::notEqual, Long);
426 // dtos
427 __ fld_d( Address(rcx, rbx, Address::times_ptr, base_offset));
428 __ push(dtos);
429 __ jmpb(Done);
430
431 __ bind(Long);
432 // ltos
433 __ movptr(rax, Address(rcx, rbx, Address::times_ptr, base_offset + 0 * wordSize));
434 NOT_LP64(__ movptr(rdx, Address(rcx, rbx, Address::times_ptr, base_offset + 1 * wordSize)));
435
436 __ push(ltos);
437
438 __ bind(Done);
439 }
440
441
442 void TemplateTable::locals_index(Register reg, int offset) {
443 __ load_unsigned_byte(reg, at_bcp(offset));
444 __ negptr(reg);
445 }
446
447
448 void TemplateTable::iload() {
449 transition(vtos, itos);
450 if (RewriteFrequentPairs) {
451 Label rewrite, done;
452
453 // get next byte
454 __ load_unsigned_byte(rbx, at_bcp(Bytecodes::length_for(Bytecodes::_iload)));
455 // if _iload, wait to rewrite to iload2. We only want to rewrite the
456 // last two iloads in a pair. Comparing against fast_iload means that
457 // the next bytecode is neither an iload or a caload, and therefore
458 // an iload pair.
459 __ cmpl(rbx, Bytecodes::_iload);
460 __ jcc(Assembler::equal, done);
461
462 __ cmpl(rbx, Bytecodes::_fast_iload);
463 __ movl(rcx, Bytecodes::_fast_iload2);
464 __ jccb(Assembler::equal, rewrite);
465
466 // if _caload, rewrite to fast_icaload
467 __ cmpl(rbx, Bytecodes::_caload);
468 __ movl(rcx, Bytecodes::_fast_icaload);
469 __ jccb(Assembler::equal, rewrite);
470
471 // rewrite so iload doesn't check again.
472 __ movl(rcx, Bytecodes::_fast_iload);
473
474 // rewrite
475 // rcx: fast bytecode
476 __ bind(rewrite);
477 patch_bytecode(Bytecodes::_iload, rcx, rbx, false);
478 __ bind(done);
479 }
480
481 // Get the local value into tos
482 locals_index(rbx);
483 __ movl(rax, iaddress(rbx));
484 }
485
486
487 void TemplateTable::fast_iload2() {
488 transition(vtos, itos);
489 locals_index(rbx);
490 __ movl(rax, iaddress(rbx));
491 __ push(itos);
492 locals_index(rbx, 3);
493 __ movl(rax, iaddress(rbx));
494 }
495
496 void TemplateTable::fast_iload() {
497 transition(vtos, itos);
498 locals_index(rbx);
499 __ movl(rax, iaddress(rbx));
500 }
501
502
503 void TemplateTable::lload() {
504 transition(vtos, ltos);
505 locals_index(rbx);
506 __ movptr(rax, laddress(rbx));
507 NOT_LP64(__ movl(rdx, haddress(rbx)));
508 }
509
510
511 void TemplateTable::fload() {
512 transition(vtos, ftos);
513 locals_index(rbx);
514 __ fld_s(faddress(rbx));
515 }
516
517
518 void TemplateTable::dload() {
519 transition(vtos, dtos);
520 locals_index(rbx);
521 __ fld_d(daddress(rbx));
522 }
523
524
525 void TemplateTable::aload() {
526 transition(vtos, atos);
527 locals_index(rbx);
528 __ movptr(rax, aaddress(rbx));
529 }
530
531
532 void TemplateTable::locals_index_wide(Register reg) {
533 __ movl(reg, at_bcp(2));
534 __ bswapl(reg);
535 __ shrl(reg, 16);
536 __ negptr(reg);
537 }
538
539
540 void TemplateTable::wide_iload() {
541 transition(vtos, itos);
542 locals_index_wide(rbx);
543 __ movl(rax, iaddress(rbx));
544 }
545
546
547 void TemplateTable::wide_lload() {
548 transition(vtos, ltos);
549 locals_index_wide(rbx);
550 __ movptr(rax, laddress(rbx));
551 NOT_LP64(__ movl(rdx, haddress(rbx)));
552 }
553
554
555 void TemplateTable::wide_fload() {
556 transition(vtos, ftos);
557 locals_index_wide(rbx);
558 __ fld_s(faddress(rbx));
559 }
560
561
562 void TemplateTable::wide_dload() {
563 transition(vtos, dtos);
564 locals_index_wide(rbx);
565 __ fld_d(daddress(rbx));
566 }
567
568
569 void TemplateTable::wide_aload() {
570 transition(vtos, atos);
571 locals_index_wide(rbx);
572 __ movptr(rax, aaddress(rbx));
573 }
574
575 void TemplateTable::index_check(Register array, Register index) {
576 // Pop ptr into array
577 __ pop_ptr(array);
578 index_check_without_pop(array, index);
579 }
580
581 void TemplateTable::index_check_without_pop(Register array, Register index) {
582 // destroys rbx,
583 // check array
584 __ null_check(array, arrayOopDesc::length_offset_in_bytes());
585 LP64_ONLY(__ movslq(index, index));
586 // check index
587 __ cmpl(index, Address(array, arrayOopDesc::length_offset_in_bytes()));
588 if (index != rbx) {
589 // ??? convention: move aberrant index into rbx, for exception message
590 assert(rbx != array, "different registers");
591 __ mov(rbx, index);
592 }
593 __ jump_cc(Assembler::aboveEqual,
594 ExternalAddress(Interpreter::_throw_ArrayIndexOutOfBoundsException_entry));
595 }
596
597
598 void TemplateTable::iaload() {
599 transition(itos, itos);
600 // rdx: array
601 index_check(rdx, rax); // kills rbx,
602 // rax,: index
603 __ movl(rax, Address(rdx, rax, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_INT)));
604 }
605
606
607 void TemplateTable::laload() {
608 transition(itos, ltos);
609 // rax,: index
610 // rdx: array
611 index_check(rdx, rax);
612 __ mov(rbx, rax);
613 // rbx,: index
614 __ movptr(rax, Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 0 * wordSize));
615 NOT_LP64(__ movl(rdx, Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 1 * wordSize)));
616 }
617
618
619 void TemplateTable::faload() {
620 transition(itos, ftos);
621 // rdx: array
622 index_check(rdx, rax); // kills rbx,
623 // rax,: index
624 __ fld_s(Address(rdx, rax, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_FLOAT)));
625 }
626
627
628 void TemplateTable::daload() {
629 transition(itos, dtos);
630 // rdx: array
631 index_check(rdx, rax); // kills rbx,
632 // rax,: index
633 __ fld_d(Address(rdx, rax, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_DOUBLE)));
634 }
635
636
637 void TemplateTable::aaload() {
638 transition(itos, atos);
639 // rdx: array
640 index_check(rdx, rax); // kills rbx,
641 // rax,: index
642 __ movptr(rax, Address(rdx, rax, Address::times_ptr, arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
643 }
644
645
646 void TemplateTable::baload() {
647 transition(itos, itos);
648 // rdx: array
649 index_check(rdx, rax); // kills rbx,
650 // rax,: index
651 // can do better code for P5 - fix this at some point
652 __ load_signed_byte(rbx, Address(rdx, rax, Address::times_1, arrayOopDesc::base_offset_in_bytes(T_BYTE)));
653 __ mov(rax, rbx);
654 }
655
656
657 void TemplateTable::caload() {
658 transition(itos, itos);
659 // rdx: array
660 index_check(rdx, rax); // kills rbx,
661 // rax,: index
662 // can do better code for P5 - may want to improve this at some point
663 __ load_unsigned_short(rbx, Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR)));
664 __ mov(rax, rbx);
665 }
666
667 // iload followed by caload frequent pair
668 void TemplateTable::fast_icaload() {
669 transition(vtos, itos);
670 // load index out of locals
671 locals_index(rbx);
672 __ movl(rax, iaddress(rbx));
673
674 // rdx: array
675 index_check(rdx, rax);
676 // rax,: index
677 __ load_unsigned_short(rbx, Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR)));
678 __ mov(rax, rbx);
679 }
680
681 void TemplateTable::saload() {
682 transition(itos, itos);
683 // rdx: array
684 index_check(rdx, rax); // kills rbx,
685 // rax,: index
686 // can do better code for P5 - may want to improve this at some point
687 __ load_signed_short(rbx, Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_SHORT)));
688 __ mov(rax, rbx);
689 }
690
691
692 void TemplateTable::iload(int n) {
693 transition(vtos, itos);
694 __ movl(rax, iaddress(n));
695 }
696
697
698 void TemplateTable::lload(int n) {
699 transition(vtos, ltos);
700 __ movptr(rax, laddress(n));
701 NOT_LP64(__ movptr(rdx, haddress(n)));
702 }
703
704
705 void TemplateTable::fload(int n) {
706 transition(vtos, ftos);
707 __ fld_s(faddress(n));
708 }
709
710
711 void TemplateTable::dload(int n) {
712 transition(vtos, dtos);
713 __ fld_d(daddress(n));
714 }
715
716
717 void TemplateTable::aload(int n) {
718 transition(vtos, atos);
719 __ movptr(rax, aaddress(n));
720 }
721
722
723 void TemplateTable::aload_0() {
724 transition(vtos, atos);
725 // According to bytecode histograms, the pairs:
726 //
727 // _aload_0, _fast_igetfield
728 // _aload_0, _fast_agetfield
729 // _aload_0, _fast_fgetfield
730 //
731 // occur frequently. If RewriteFrequentPairs is set, the (slow) _aload_0
732 // bytecode checks if the next bytecode is either _fast_igetfield,
733 // _fast_agetfield or _fast_fgetfield and then rewrites the
734 // current bytecode into a pair bytecode; otherwise it rewrites the current
735 // bytecode into _fast_aload_0 that doesn't do the pair check anymore.
736 //
737 // Note: If the next bytecode is _getfield, the rewrite must be delayed,
738 // otherwise we may miss an opportunity for a pair.
739 //
740 // Also rewrite frequent pairs
741 // aload_0, aload_1
742 // aload_0, iload_1
743 // These bytecodes with a small amount of code are most profitable to rewrite
744 if (RewriteFrequentPairs) {
745 Label rewrite, done;
746 // get next byte
747 __ load_unsigned_byte(rbx, at_bcp(Bytecodes::length_for(Bytecodes::_aload_0)));
748
749 // do actual aload_0
750 aload(0);
751
752 // if _getfield then wait with rewrite
753 __ cmpl(rbx, Bytecodes::_getfield);
754 __ jcc(Assembler::equal, done);
755
756 // if _igetfield then reqrite to _fast_iaccess_0
757 assert(Bytecodes::java_code(Bytecodes::_fast_iaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
758 __ cmpl(rbx, Bytecodes::_fast_igetfield);
759 __ movl(rcx, Bytecodes::_fast_iaccess_0);
760 __ jccb(Assembler::equal, rewrite);
761
762 // if _agetfield then reqrite to _fast_aaccess_0
763 assert(Bytecodes::java_code(Bytecodes::_fast_aaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
764 __ cmpl(rbx, Bytecodes::_fast_agetfield);
765 __ movl(rcx, Bytecodes::_fast_aaccess_0);
766 __ jccb(Assembler::equal, rewrite);
767
768 // if _fgetfield then reqrite to _fast_faccess_0
769 assert(Bytecodes::java_code(Bytecodes::_fast_faccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
770 __ cmpl(rbx, Bytecodes::_fast_fgetfield);
771 __ movl(rcx, Bytecodes::_fast_faccess_0);
772 __ jccb(Assembler::equal, rewrite);
773
774 // else rewrite to _fast_aload0
775 assert(Bytecodes::java_code(Bytecodes::_fast_aload_0) == Bytecodes::_aload_0, "fix bytecode definition");
776 __ movl(rcx, Bytecodes::_fast_aload_0);
777
778 // rewrite
779 // rcx: fast bytecode
780 __ bind(rewrite);
781 patch_bytecode(Bytecodes::_aload_0, rcx, rbx, false);
782
783 __ bind(done);
784 } else {
785 aload(0);
786 }
787 }
788
789 void TemplateTable::istore() {
790 transition(itos, vtos);
791 locals_index(rbx);
792 __ movl(iaddress(rbx), rax);
793 }
794
795
796 void TemplateTable::lstore() {
797 transition(ltos, vtos);
798 locals_index(rbx);
799 __ movptr(laddress(rbx), rax);
800 NOT_LP64(__ movptr(haddress(rbx), rdx));
801 }
802
803
804 void TemplateTable::fstore() {
805 transition(ftos, vtos);
806 locals_index(rbx);
807 __ fstp_s(faddress(rbx));
808 }
809
810
811 void TemplateTable::dstore() {
812 transition(dtos, vtos);
813 locals_index(rbx);
814 __ fstp_d(daddress(rbx));
815 }
816
817
818 void TemplateTable::astore() {
819 transition(vtos, vtos);
820 __ pop_ptr(rax);
821 locals_index(rbx);
822 __ movptr(aaddress(rbx), rax);
823 }
824
825
826 void TemplateTable::wide_istore() {
827 transition(vtos, vtos);
828 __ pop_i(rax);
829 locals_index_wide(rbx);
830 __ movl(iaddress(rbx), rax);
831 }
832
833
834 void TemplateTable::wide_lstore() {
835 transition(vtos, vtos);
836 __ pop_l(rax, rdx);
837 locals_index_wide(rbx);
838 __ movptr(laddress(rbx), rax);
839 NOT_LP64(__ movl(haddress(rbx), rdx));
840 }
841
842
843 void TemplateTable::wide_fstore() {
844 wide_istore();
845 }
846
847
848 void TemplateTable::wide_dstore() {
849 wide_lstore();
850 }
851
852
853 void TemplateTable::wide_astore() {
854 transition(vtos, vtos);
855 __ pop_ptr(rax);
856 locals_index_wide(rbx);
857 __ movptr(aaddress(rbx), rax);
858 }
859
860
861 void TemplateTable::iastore() {
862 transition(itos, vtos);
863 __ pop_i(rbx);
864 // rax,: value
865 // rdx: array
866 index_check(rdx, rbx); // prefer index in rbx,
867 // rbx,: index
868 __ movl(Address(rdx, rbx, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_INT)), rax);
869 }
870
871
872 void TemplateTable::lastore() {
873 transition(ltos, vtos);
874 __ pop_i(rbx);
875 // rax,: low(value)
876 // rcx: array
877 // rdx: high(value)
878 index_check(rcx, rbx); // prefer index in rbx,
879 // rbx,: index
880 __ movptr(Address(rcx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 0 * wordSize), rax);
881 NOT_LP64(__ movl(Address(rcx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 1 * wordSize), rdx));
882 }
883
884
885 void TemplateTable::fastore() {
886 transition(ftos, vtos);
887 __ pop_i(rbx);
888 // rdx: array
889 // st0: value
890 index_check(rdx, rbx); // prefer index in rbx,
891 // rbx,: index
892 __ fstp_s(Address(rdx, rbx, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_FLOAT)));
893 }
894
895
896 void TemplateTable::dastore() {
897 transition(dtos, vtos);
898 __ pop_i(rbx);
899 // rdx: array
900 // st0: value
901 index_check(rdx, rbx); // prefer index in rbx,
902 // rbx,: index
903 __ fstp_d(Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_DOUBLE)));
904 }
905
906
907 void TemplateTable::aastore() {
908 Label is_null, ok_is_subtype, done;
909 transition(vtos, vtos);
910 // stack: ..., array, index, value
911 __ movptr(rax, at_tos()); // Value
912 __ movl(rcx, at_tos_p1()); // Index
913 __ movptr(rdx, at_tos_p2()); // Array
914
915 Address element_address(rdx, rcx, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_OBJECT));
916 index_check_without_pop(rdx, rcx); // kills rbx,
917 // do array store check - check for NULL value first
918 __ testptr(rax, rax);
919 __ jcc(Assembler::zero, is_null);
920
921 // Move subklass into EBX
922 __ movptr(rbx, Address(rax, oopDesc::klass_offset_in_bytes()));
923 // Move superklass into EAX
924 __ movptr(rax, Address(rdx, oopDesc::klass_offset_in_bytes()));
925 __ movptr(rax, Address(rax, sizeof(oopDesc) + objArrayKlass::element_klass_offset_in_bytes()));
926 // Compress array+index*wordSize+12 into a single register. Frees ECX.
927 __ lea(rdx, element_address);
928
929 // Generate subtype check. Blows ECX. Resets EDI to locals.
930 // Superklass in EAX. Subklass in EBX.
931 __ gen_subtype_check( rbx, ok_is_subtype );
932
933 // Come here on failure
934 // object is at TOS
935 __ jump(ExternalAddress(Interpreter::_throw_ArrayStoreException_entry));
936
937 // Come here on success
938 __ bind(ok_is_subtype);
939
940 // Get the value to store
941 __ movptr(rax, at_rsp());
942 // and store it with appropriate barrier
943 do_oop_store(_masm, Address(rdx, 0), rax, _bs->kind(), true);
944
945 __ jmp(done);
946
947 // Have a NULL in EAX, EDX=array, ECX=index. Store NULL at ary[idx]
948 __ bind(is_null);
949 __ profile_null_seen(rbx);
950
951 // Store NULL, (noreg means NULL to do_oop_store)
952 do_oop_store(_masm, element_address, noreg, _bs->kind(), true);
953
954 // Pop stack arguments
955 __ bind(done);
956 __ addptr(rsp, 3 * Interpreter::stackElementSize);
957 }
958
959
960 void TemplateTable::bastore() {
961 transition(itos, vtos);
962 __ pop_i(rbx);
963 // rax,: value
964 // rdx: array
965 index_check(rdx, rbx); // prefer index in rbx,
966 // rbx,: index
967 __ movb(Address(rdx, rbx, Address::times_1, arrayOopDesc::base_offset_in_bytes(T_BYTE)), rax);
968 }
969
970
971 void TemplateTable::castore() {
972 transition(itos, vtos);
973 __ pop_i(rbx);
974 // rax,: value
975 // rdx: array
976 index_check(rdx, rbx); // prefer index in rbx,
977 // rbx,: index
978 __ movw(Address(rdx, rbx, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR)), rax);
979 }
980
981
982 void TemplateTable::sastore() {
983 castore();
984 }
985
986
987 void TemplateTable::istore(int n) {
988 transition(itos, vtos);
989 __ movl(iaddress(n), rax);
990 }
991
992
993 void TemplateTable::lstore(int n) {
994 transition(ltos, vtos);
995 __ movptr(laddress(n), rax);
996 NOT_LP64(__ movptr(haddress(n), rdx));
997 }
998
999
1000 void TemplateTable::fstore(int n) {
1001 transition(ftos, vtos);
1002 __ fstp_s(faddress(n));
1003 }
1004
1005
1006 void TemplateTable::dstore(int n) {
1007 transition(dtos, vtos);
1008 __ fstp_d(daddress(n));
1009 }
1010
1011
1012 void TemplateTable::astore(int n) {
1013 transition(vtos, vtos);
1014 __ pop_ptr(rax);
1015 __ movptr(aaddress(n), rax);
1016 }
1017
1018
1019 void TemplateTable::pop() {
1020 transition(vtos, vtos);
1021 __ addptr(rsp, Interpreter::stackElementSize);
1022 }
1023
1024
1025 void TemplateTable::pop2() {
1026 transition(vtos, vtos);
1027 __ addptr(rsp, 2*Interpreter::stackElementSize);
1028 }
1029
1030
1031 void TemplateTable::dup() {
1032 transition(vtos, vtos);
1033 // stack: ..., a
1034 __ load_ptr(0, rax);
1035 __ push_ptr(rax);
1036 // stack: ..., a, a
1037 }
1038
1039
1040 void TemplateTable::dup_x1() {
1041 transition(vtos, vtos);
1042 // stack: ..., a, b
1043 __ load_ptr( 0, rax); // load b
1044 __ load_ptr( 1, rcx); // load a
1045 __ store_ptr(1, rax); // store b
1046 __ store_ptr(0, rcx); // store a
1047 __ push_ptr(rax); // push b
1048 // stack: ..., b, a, b
1049 }
1050
1051
1052 void TemplateTable::dup_x2() {
1053 transition(vtos, vtos);
1054 // stack: ..., a, b, c
1055 __ load_ptr( 0, rax); // load c
1056 __ load_ptr( 2, rcx); // load a
1057 __ store_ptr(2, rax); // store c in a
1058 __ push_ptr(rax); // push c
1059 // stack: ..., c, b, c, c
1060 __ load_ptr( 2, rax); // load b
1061 __ store_ptr(2, rcx); // store a in b
1062 // stack: ..., c, a, c, c
1063 __ store_ptr(1, rax); // store b in c
1064 // stack: ..., c, a, b, c
1065 }
1066
1067
1068 void TemplateTable::dup2() {
1069 transition(vtos, vtos);
1070 // stack: ..., a, b
1071 __ load_ptr(1, rax); // load a
1072 __ push_ptr(rax); // push a
1073 __ load_ptr(1, rax); // load b
1074 __ push_ptr(rax); // push b
1075 // stack: ..., a, b, a, b
1076 }
1077
1078
1079 void TemplateTable::dup2_x1() {
1080 transition(vtos, vtos);
1081 // stack: ..., a, b, c
1082 __ load_ptr( 0, rcx); // load c
1083 __ load_ptr( 1, rax); // load b
1084 __ push_ptr(rax); // push b
1085 __ push_ptr(rcx); // push c
1086 // stack: ..., a, b, c, b, c
1087 __ store_ptr(3, rcx); // store c in b
1088 // stack: ..., a, c, c, b, c
1089 __ load_ptr( 4, rcx); // load a
1090 __ store_ptr(2, rcx); // store a in 2nd c
1091 // stack: ..., a, c, a, b, c
1092 __ store_ptr(4, rax); // store b in a
1093 // stack: ..., b, c, a, b, c
1094 // stack: ..., b, c, a, b, c
1095 }
1096
1097
1098 void TemplateTable::dup2_x2() {
1099 transition(vtos, vtos);
1100 // stack: ..., a, b, c, d
1101 __ load_ptr( 0, rcx); // load d
1102 __ load_ptr( 1, rax); // load c
1103 __ push_ptr(rax); // push c
1104 __ push_ptr(rcx); // push d
1105 // stack: ..., a, b, c, d, c, d
1106 __ load_ptr( 4, rax); // load b
1107 __ store_ptr(2, rax); // store b in d
1108 __ store_ptr(4, rcx); // store d in b
1109 // stack: ..., a, d, c, b, c, d
1110 __ load_ptr( 5, rcx); // load a
1111 __ load_ptr( 3, rax); // load c
1112 __ store_ptr(3, rcx); // store a in c
1113 __ store_ptr(5, rax); // store c in a
1114 // stack: ..., c, d, a, b, c, d
1115 // stack: ..., c, d, a, b, c, d
1116 }
1117
1118
1119 void TemplateTable::swap() {
1120 transition(vtos, vtos);
1121 // stack: ..., a, b
1122 __ load_ptr( 1, rcx); // load a
1123 __ load_ptr( 0, rax); // load b
1124 __ store_ptr(0, rcx); // store a in b
1125 __ store_ptr(1, rax); // store b in a
1126 // stack: ..., b, a
1127 }
1128
1129
1130 void TemplateTable::iop2(Operation op) {
1131 transition(itos, itos);
1132 switch (op) {
1133 case add : __ pop_i(rdx); __ addl (rax, rdx); break;
1134 case sub : __ mov(rdx, rax); __ pop_i(rax); __ subl (rax, rdx); break;
1135 case mul : __ pop_i(rdx); __ imull(rax, rdx); break;
1136 case _and : __ pop_i(rdx); __ andl (rax, rdx); break;
1137 case _or : __ pop_i(rdx); __ orl (rax, rdx); break;
1138 case _xor : __ pop_i(rdx); __ xorl (rax, rdx); break;
1139 case shl : __ mov(rcx, rax); __ pop_i(rax); __ shll (rax); break; // implicit masking of lower 5 bits by Intel shift instr.
1140 case shr : __ mov(rcx, rax); __ pop_i(rax); __ sarl (rax); break; // implicit masking of lower 5 bits by Intel shift instr.
1141 case ushr : __ mov(rcx, rax); __ pop_i(rax); __ shrl (rax); break; // implicit masking of lower 5 bits by Intel shift instr.
1142 default : ShouldNotReachHere();
1143 }
1144 }
1145
1146
1147 void TemplateTable::lop2(Operation op) {
1148 transition(ltos, ltos);
1149 __ pop_l(rbx, rcx);
1150 switch (op) {
1151 case add : __ addl(rax, rbx); __ adcl(rdx, rcx); break;
1152 case sub : __ subl(rbx, rax); __ sbbl(rcx, rdx);
1153 __ mov (rax, rbx); __ mov (rdx, rcx); break;
1154 case _and : __ andl(rax, rbx); __ andl(rdx, rcx); break;
1155 case _or : __ orl (rax, rbx); __ orl (rdx, rcx); break;
1156 case _xor : __ xorl(rax, rbx); __ xorl(rdx, rcx); break;
1157 default : ShouldNotReachHere();
1158 }
1159 }
1160
1161
1162 void TemplateTable::idiv() {
1163 transition(itos, itos);
1164 __ mov(rcx, rax);
1165 __ pop_i(rax);
1166 // Note: could xor rax, and rcx and compare with (-1 ^ min_int). If
1167 // they are not equal, one could do a normal division (no correction
1168 // needed), which may speed up this implementation for the common case.
1169 // (see also JVM spec., p.243 & p.271)
1170 __ corrected_idivl(rcx);
1171 }
1172
1173
1174 void TemplateTable::irem() {
1175 transition(itos, itos);
1176 __ mov(rcx, rax);
1177 __ pop_i(rax);
1178 // Note: could xor rax, and rcx and compare with (-1 ^ min_int). If
1179 // they are not equal, one could do a normal division (no correction
1180 // needed), which may speed up this implementation for the common case.
1181 // (see also JVM spec., p.243 & p.271)
1182 __ corrected_idivl(rcx);
1183 __ mov(rax, rdx);
1184 }
1185
1186
1187 void TemplateTable::lmul() {
1188 transition(ltos, ltos);
1189 __ pop_l(rbx, rcx);
1190 __ push(rcx); __ push(rbx);
1191 __ push(rdx); __ push(rax);
1192 __ lmul(2 * wordSize, 0);
1193 __ addptr(rsp, 4 * wordSize); // take off temporaries
1194 }
1195
1196
1197 void TemplateTable::ldiv() {
1198 transition(ltos, ltos);
1199 __ pop_l(rbx, rcx);
1200 __ push(rcx); __ push(rbx);
1201 __ push(rdx); __ push(rax);
1202 // check if y = 0
1203 __ orl(rax, rdx);
1204 __ jump_cc(Assembler::zero,
1205 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1206 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::ldiv));
1207 __ addptr(rsp, 4 * wordSize); // take off temporaries
1208 }
1209
1210
1211 void TemplateTable::lrem() {
1212 transition(ltos, ltos);
1213 __ pop_l(rbx, rcx);
1214 __ push(rcx); __ push(rbx);
1215 __ push(rdx); __ push(rax);
1216 // check if y = 0
1217 __ orl(rax, rdx);
1218 __ jump_cc(Assembler::zero,
1219 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1220 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::lrem));
1221 __ addptr(rsp, 4 * wordSize);
1222 }
1223
1224
1225 void TemplateTable::lshl() {
1226 transition(itos, ltos);
1227 __ movl(rcx, rax); // get shift count
1228 __ pop_l(rax, rdx); // get shift value
1229 __ lshl(rdx, rax);
1230 }
1231
1232
1233 void TemplateTable::lshr() {
1234 transition(itos, ltos);
1235 __ mov(rcx, rax); // get shift count
1236 __ pop_l(rax, rdx); // get shift value
1237 __ lshr(rdx, rax, true);
1238 }
1239
1240
1241 void TemplateTable::lushr() {
1242 transition(itos, ltos);
1243 __ mov(rcx, rax); // get shift count
1244 __ pop_l(rax, rdx); // get shift value
1245 __ lshr(rdx, rax);
1246 }
1247
1248
1249 void TemplateTable::fop2(Operation op) {
1250 transition(ftos, ftos);
1251 switch (op) {
1252 case add: __ fadd_s (at_rsp()); break;
1253 case sub: __ fsubr_s(at_rsp()); break;
1254 case mul: __ fmul_s (at_rsp()); break;
1255 case div: __ fdivr_s(at_rsp()); break;
1256 case rem: __ fld_s (at_rsp()); __ fremr(rax); break;
1257 default : ShouldNotReachHere();
1258 }
1259 __ f2ieee();
1260 __ pop(rax); // pop float thing off
1261 }
1262
1263
1264 void TemplateTable::dop2(Operation op) {
1265 transition(dtos, dtos);
1266
1267 switch (op) {
1268 case add: __ fadd_d (at_rsp()); break;
1269 case sub: __ fsubr_d(at_rsp()); break;
1270 case mul: {
1271 Label L_strict;
1272 Label L_join;
1273 const Address access_flags (rcx, methodOopDesc::access_flags_offset());
1274 __ get_method(rcx);
1275 __ movl(rcx, access_flags);
1276 __ testl(rcx, JVM_ACC_STRICT);
1277 __ jccb(Assembler::notZero, L_strict);
1278 __ fmul_d (at_rsp());
1279 __ jmpb(L_join);
1280 __ bind(L_strict);
1281 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias1()));
1282 __ fmulp();
1283 __ fmul_d (at_rsp());
1284 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias2()));
1285 __ fmulp();
1286 __ bind(L_join);
1287 break;
1288 }
1289 case div: {
1290 Label L_strict;
1291 Label L_join;
1292 const Address access_flags (rcx, methodOopDesc::access_flags_offset());
1293 __ get_method(rcx);
1294 __ movl(rcx, access_flags);
1295 __ testl(rcx, JVM_ACC_STRICT);
1296 __ jccb(Assembler::notZero, L_strict);
1297 __ fdivr_d(at_rsp());
1298 __ jmp(L_join);
1299 __ bind(L_strict);
1300 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias1()));
1301 __ fmul_d (at_rsp());
1302 __ fdivrp();
1303 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias2()));
1304 __ fmulp();
1305 __ bind(L_join);
1306 break;
1307 }
1308 case rem: __ fld_d (at_rsp()); __ fremr(rax); break;
1309 default : ShouldNotReachHere();
1310 }
1311 __ d2ieee();
1312 // Pop double precision number from rsp.
1313 __ pop(rax);
1314 __ pop(rdx);
1315 }
1316
1317
1318 void TemplateTable::ineg() {
1319 transition(itos, itos);
1320 __ negl(rax);
1321 }
1322
1323
1324 void TemplateTable::lneg() {
1325 transition(ltos, ltos);
1326 __ lneg(rdx, rax);
1327 }
1328
1329
1330 void TemplateTable::fneg() {
1331 transition(ftos, ftos);
1332 __ fchs();
1333 }
1334
1335
1336 void TemplateTable::dneg() {
1337 transition(dtos, dtos);
1338 __ fchs();
1339 }
1340
1341
1342 void TemplateTable::iinc() {
1343 transition(vtos, vtos);
1344 __ load_signed_byte(rdx, at_bcp(2)); // get constant
1345 locals_index(rbx);
1346 __ addl(iaddress(rbx), rdx);
1347 }
1348
1349
1350 void TemplateTable::wide_iinc() {
1351 transition(vtos, vtos);
1352 __ movl(rdx, at_bcp(4)); // get constant
1353 locals_index_wide(rbx);
1354 __ bswapl(rdx); // swap bytes & sign-extend constant
1355 __ sarl(rdx, 16);
1356 __ addl(iaddress(rbx), rdx);
1357 // Note: should probably use only one movl to get both
1358 // the index and the constant -> fix this
1359 }
1360
1361
1362 void TemplateTable::convert() {
1363 // Checking
1364 #ifdef ASSERT
1365 { TosState tos_in = ilgl;
1366 TosState tos_out = ilgl;
1367 switch (bytecode()) {
1368 case Bytecodes::_i2l: // fall through
1369 case Bytecodes::_i2f: // fall through
1370 case Bytecodes::_i2d: // fall through
1371 case Bytecodes::_i2b: // fall through
1372 case Bytecodes::_i2c: // fall through
1373 case Bytecodes::_i2s: tos_in = itos; break;
1374 case Bytecodes::_l2i: // fall through
1375 case Bytecodes::_l2f: // fall through
1376 case Bytecodes::_l2d: tos_in = ltos; break;
1377 case Bytecodes::_f2i: // fall through
1378 case Bytecodes::_f2l: // fall through
1379 case Bytecodes::_f2d: tos_in = ftos; break;
1380 case Bytecodes::_d2i: // fall through
1381 case Bytecodes::_d2l: // fall through
1382 case Bytecodes::_d2f: tos_in = dtos; break;
1383 default : ShouldNotReachHere();
1384 }
1385 switch (bytecode()) {
1386 case Bytecodes::_l2i: // fall through
1387 case Bytecodes::_f2i: // fall through
1388 case Bytecodes::_d2i: // fall through
1389 case Bytecodes::_i2b: // fall through
1390 case Bytecodes::_i2c: // fall through
1391 case Bytecodes::_i2s: tos_out = itos; break;
1392 case Bytecodes::_i2l: // fall through
1393 case Bytecodes::_f2l: // fall through
1394 case Bytecodes::_d2l: tos_out = ltos; break;
1395 case Bytecodes::_i2f: // fall through
1396 case Bytecodes::_l2f: // fall through
1397 case Bytecodes::_d2f: tos_out = ftos; break;
1398 case Bytecodes::_i2d: // fall through
1399 case Bytecodes::_l2d: // fall through
1400 case Bytecodes::_f2d: tos_out = dtos; break;
1401 default : ShouldNotReachHere();
1402 }
1403 transition(tos_in, tos_out);
1404 }
1405 #endif // ASSERT
1406
1407 // Conversion
1408 // (Note: use push(rcx)/pop(rcx) for 1/2-word stack-ptr manipulation)
1409 switch (bytecode()) {
1410 case Bytecodes::_i2l:
1411 __ extend_sign(rdx, rax);
1412 break;
1413 case Bytecodes::_i2f:
1414 __ push(rax); // store int on tos
1415 __ fild_s(at_rsp()); // load int to ST0
1416 __ f2ieee(); // truncate to float size
1417 __ pop(rcx); // adjust rsp
1418 break;
1419 case Bytecodes::_i2d:
1420 __ push(rax); // add one slot for d2ieee()
1421 __ push(rax); // store int on tos
1422 __ fild_s(at_rsp()); // load int to ST0
1423 __ d2ieee(); // truncate to double size
1424 __ pop(rcx); // adjust rsp
1425 __ pop(rcx);
1426 break;
1427 case Bytecodes::_i2b:
1428 __ shll(rax, 24); // truncate upper 24 bits
1429 __ sarl(rax, 24); // and sign-extend byte
1430 LP64_ONLY(__ movsbl(rax, rax));
1431 break;
1432 case Bytecodes::_i2c:
1433 __ andl(rax, 0xFFFF); // truncate upper 16 bits
1434 LP64_ONLY(__ movzwl(rax, rax));
1435 break;
1436 case Bytecodes::_i2s:
1437 __ shll(rax, 16); // truncate upper 16 bits
1438 __ sarl(rax, 16); // and sign-extend short
1439 LP64_ONLY(__ movswl(rax, rax));
1440 break;
1441 case Bytecodes::_l2i:
1442 /* nothing to do */
1443 break;
1444 case Bytecodes::_l2f:
1445 __ push(rdx); // store long on tos
1446 __ push(rax);
1447 __ fild_d(at_rsp()); // load long to ST0
1448 __ f2ieee(); // truncate to float size
1449 __ pop(rcx); // adjust rsp
1450 __ pop(rcx);
1451 break;
1452 case Bytecodes::_l2d:
1453 __ push(rdx); // store long on tos
1454 __ push(rax);
1455 __ fild_d(at_rsp()); // load long to ST0
1456 __ d2ieee(); // truncate to double size
1457 __ pop(rcx); // adjust rsp
1458 __ pop(rcx);
1459 break;
1460 case Bytecodes::_f2i:
1461 __ push(rcx); // reserve space for argument
1462 __ fstp_s(at_rsp()); // pass float argument on stack
1463 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i), 1);
1464 break;
1465 case Bytecodes::_f2l:
1466 __ push(rcx); // reserve space for argument
1467 __ fstp_s(at_rsp()); // pass float argument on stack
1468 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l), 1);
1469 break;
1470 case Bytecodes::_f2d:
1471 /* nothing to do */
1472 break;
1473 case Bytecodes::_d2i:
1474 __ push(rcx); // reserve space for argument
1475 __ push(rcx);
1476 __ fstp_d(at_rsp()); // pass double argument on stack
1477 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2i), 2);
1478 break;
1479 case Bytecodes::_d2l:
1480 __ push(rcx); // reserve space for argument
1481 __ push(rcx);
1482 __ fstp_d(at_rsp()); // pass double argument on stack
1483 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2l), 2);
1484 break;
1485 case Bytecodes::_d2f:
1486 __ push(rcx); // reserve space for f2ieee()
1487 __ f2ieee(); // truncate to float size
1488 __ pop(rcx); // adjust rsp
1489 break;
1490 default :
1491 ShouldNotReachHere();
1492 }
1493 }
1494
1495
1496 void TemplateTable::lcmp() {
1497 transition(ltos, itos);
1498 // y = rdx:rax
1499 __ pop_l(rbx, rcx); // get x = rcx:rbx
1500 __ lcmp2int(rcx, rbx, rdx, rax);// rcx := cmp(x, y)
1501 __ mov(rax, rcx);
1502 }
1503
1504
1505 void TemplateTable::float_cmp(bool is_float, int unordered_result) {
1506 if (is_float) {
1507 __ fld_s(at_rsp());
1508 } else {
1509 __ fld_d(at_rsp());
1510 __ pop(rdx);
1511 }
1512 __ pop(rcx);
1513 __ fcmp2int(rax, unordered_result < 0);
1514 }
1515
1516
1517 void TemplateTable::branch(bool is_jsr, bool is_wide) {
1518 __ get_method(rcx); // ECX holds method
1519 __ profile_taken_branch(rax,rbx); // EAX holds updated MDP, EBX holds bumped taken count
1520
1521 const ByteSize be_offset = methodOopDesc::backedge_counter_offset() + InvocationCounter::counter_offset();
1522 const ByteSize inv_offset = methodOopDesc::invocation_counter_offset() + InvocationCounter::counter_offset();
1523 const int method_offset = frame::interpreter_frame_method_offset * wordSize;
1524
1525 // Load up EDX with the branch displacement
1526 __ movl(rdx, at_bcp(1));
1527 __ bswapl(rdx);
1528 if (!is_wide) __ sarl(rdx, 16);
1529 LP64_ONLY(__ movslq(rdx, rdx));
1530
1531
1532 // Handle all the JSR stuff here, then exit.
1533 // It's much shorter and cleaner than intermingling with the
1534 // non-JSR normal-branch stuff occurring below.
1535 if (is_jsr) {
1536 // Pre-load the next target bytecode into EBX
1537 __ load_unsigned_byte(rbx, Address(rsi, rdx, Address::times_1, 0));
1538
1539 // compute return address as bci in rax,
1540 __ lea(rax, at_bcp((is_wide ? 5 : 3) - in_bytes(constMethodOopDesc::codes_offset())));
1541 __ subptr(rax, Address(rcx, methodOopDesc::const_offset()));
1542 // Adjust the bcp in RSI by the displacement in EDX
1543 __ addptr(rsi, rdx);
1544 // Push return address
1545 __ push_i(rax);
1546 // jsr returns vtos
1547 __ dispatch_only_noverify(vtos);
1548 return;
1549 }
1550
1551 // Normal (non-jsr) branch handling
1552
1553 // Adjust the bcp in RSI by the displacement in EDX
1554 __ addptr(rsi, rdx);
1555
1556 assert(UseLoopCounter || !UseOnStackReplacement, "on-stack-replacement requires loop counters");
1557 Label backedge_counter_overflow;
1558 Label profile_method;
1559 Label dispatch;
1560 if (UseLoopCounter) {
1561 // increment backedge counter for backward branches
1562 // rax,: MDO
1563 // rbx,: MDO bumped taken-count
1564 // rcx: method
1565 // rdx: target offset
1566 // rsi: target bcp
1567 // rdi: locals pointer
1568 __ testl(rdx, rdx); // check if forward or backward branch
1569 __ jcc(Assembler::positive, dispatch); // count only if backward branch
1570
1571 if (TieredCompilation) {
1572 Label no_mdo;
1573 int increment = InvocationCounter::count_increment;
1574 int mask = ((1 << Tier0BackedgeNotifyFreqLog) - 1) << InvocationCounter::count_shift;
1575 if (ProfileInterpreter) {
1576 // Are we profiling?
1577 __ movptr(rbx, Address(rcx, in_bytes(methodOopDesc::method_data_offset())));
1578 __ testptr(rbx, rbx);
1579 __ jccb(Assembler::zero, no_mdo);
1580 // Increment the MDO backedge counter
1581 const Address mdo_backedge_counter(rbx, in_bytes(methodDataOopDesc::backedge_counter_offset()) +
1582 in_bytes(InvocationCounter::counter_offset()));
1583 __ increment_mask_and_jump(mdo_backedge_counter, increment, mask,
1584 rax, false, Assembler::zero, &backedge_counter_overflow);
1585 __ jmp(dispatch);
1586 }
1587 __ bind(no_mdo);
1588 // Increment backedge counter in methodOop
1589 __ increment_mask_and_jump(Address(rcx, be_offset), increment, mask,
1590 rax, false, Assembler::zero, &backedge_counter_overflow);
1591 } else {
1592 // increment counter
1593 __ movl(rax, Address(rcx, be_offset)); // load backedge counter
1594 __ incrementl(rax, InvocationCounter::count_increment); // increment counter
1595 __ movl(Address(rcx, be_offset), rax); // store counter
1596
1597 __ movl(rax, Address(rcx, inv_offset)); // load invocation counter
1598 __ andl(rax, InvocationCounter::count_mask_value); // and the status bits
1599 __ addl(rax, Address(rcx, be_offset)); // add both counters
1600
1601 if (ProfileInterpreter) {
1602 // Test to see if we should create a method data oop
1603 __ cmp32(rax,
1604 ExternalAddress((address) &InvocationCounter::InterpreterProfileLimit));
1605 __ jcc(Assembler::less, dispatch);
1606
1607 // if no method data exists, go to profile method
1608 __ test_method_data_pointer(rax, profile_method);
1609
1610 if (UseOnStackReplacement) {
1611 // check for overflow against rbx, which is the MDO taken count
1612 __ cmp32(rbx,
1613 ExternalAddress((address) &InvocationCounter::InterpreterBackwardBranchLimit));
1614 __ jcc(Assembler::below, dispatch);
1615
1616 // When ProfileInterpreter is on, the backedge_count comes from the
1617 // methodDataOop, which value does not get reset on the call to
1618 // frequency_counter_overflow(). To avoid excessive calls to the overflow
1619 // routine while the method is being compiled, add a second test to make
1620 // sure the overflow function is called only once every overflow_frequency.
1621 const int overflow_frequency = 1024;
1622 __ andptr(rbx, overflow_frequency-1);
1623 __ jcc(Assembler::zero, backedge_counter_overflow);
1624 }
1625 } else {
1626 if (UseOnStackReplacement) {
1627 // check for overflow against rax, which is the sum of the counters
1628 __ cmp32(rax,
1629 ExternalAddress((address) &InvocationCounter::InterpreterBackwardBranchLimit));
1630 __ jcc(Assembler::aboveEqual, backedge_counter_overflow);
1631
1632 }
1633 }
1634 }
1635 __ bind(dispatch);
1636 }
1637
1638 // Pre-load the next target bytecode into EBX
1639 __ load_unsigned_byte(rbx, Address(rsi, 0));
1640
1641 // continue with the bytecode @ target
1642 // rax,: return bci for jsr's, unused otherwise
1643 // rbx,: target bytecode
1644 // rsi: target bcp
1645 __ dispatch_only(vtos);
1646
1647 if (UseLoopCounter) {
1648 if (ProfileInterpreter) {
1649 // Out-of-line code to allocate method data oop.
1650 __ bind(profile_method);
1651 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method), rsi);
1652 __ load_unsigned_byte(rbx, Address(rsi, 0)); // restore target bytecode
1653 __ movptr(rcx, Address(rbp, method_offset));
1654 __ movptr(rcx, Address(rcx, in_bytes(methodOopDesc::method_data_offset())));
1655 __ movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), rcx);
1656 __ test_method_data_pointer(rcx, dispatch);
1657 // offset non-null mdp by MDO::data_offset() + IR::profile_method()
1658 __ addptr(rcx, in_bytes(methodDataOopDesc::data_offset()));
1659 __ addptr(rcx, rax);
1660 __ movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), rcx);
1661 __ jmp(dispatch);
1662 }
1663
1664 if (UseOnStackReplacement) {
1665
1666 // invocation counter overflow
1667 __ bind(backedge_counter_overflow);
1668 __ negptr(rdx);
1669 __ addptr(rdx, rsi); // branch bcp
1670 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), rdx);
1671 __ load_unsigned_byte(rbx, Address(rsi, 0)); // restore target bytecode
1672
1673 // rax,: osr nmethod (osr ok) or NULL (osr not possible)
1674 // rbx,: target bytecode
1675 // rdx: scratch
1676 // rdi: locals pointer
1677 // rsi: bcp
1678 __ testptr(rax, rax); // test result
1679 __ jcc(Assembler::zero, dispatch); // no osr if null
1680 // nmethod may have been invalidated (VM may block upon call_VM return)
1681 __ movl(rcx, Address(rax, nmethod::entry_bci_offset()));
1682 __ cmpl(rcx, InvalidOSREntryBci);
1683 __ jcc(Assembler::equal, dispatch);
1684
1685 // We have the address of an on stack replacement routine in rax,
1686 // We need to prepare to execute the OSR method. First we must
1687 // migrate the locals and monitors off of the stack.
1688
1689 __ mov(rbx, rax); // save the nmethod
1690
1691 const Register thread = rcx;
1692 __ get_thread(thread);
1693 call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin));
1694 // rax, is OSR buffer, move it to expected parameter location
1695 __ mov(rcx, rax);
1696
1697 // pop the interpreter frame
1698 __ movptr(rdx, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); // get sender sp
1699 __ leave(); // remove frame anchor
1700 __ pop(rdi); // get return address
1701 __ mov(rsp, rdx); // set sp to sender sp
1702
1703
1704 Label skip;
1705 Label chkint;
1706
1707 // The interpreter frame we have removed may be returning to
1708 // either the callstub or the interpreter. Since we will
1709 // now be returning from a compiled (OSR) nmethod we must
1710 // adjust the return to the return were it can handler compiled
1711 // results and clean the fpu stack. This is very similar to
1712 // what a i2c adapter must do.
1713
1714 // Are we returning to the call stub?
1715
1716 __ cmp32(rdi, ExternalAddress(StubRoutines::_call_stub_return_address));
1717 __ jcc(Assembler::notEqual, chkint);
1718
1719 // yes adjust to the specialized call stub return.
1720 assert(StubRoutines::x86::get_call_stub_compiled_return() != NULL, "must be set");
1721 __ lea(rdi, ExternalAddress(StubRoutines::x86::get_call_stub_compiled_return()));
1722 __ jmp(skip);
1723
1724 __ bind(chkint);
1725
1726 // Are we returning to the interpreter? Look for sentinel
1727
1728 __ cmpl(Address(rdi, -2*wordSize), Interpreter::return_sentinel);
1729 __ jcc(Assembler::notEqual, skip);
1730
1731 // Adjust to compiled return back to interpreter
1732
1733 __ movptr(rdi, Address(rdi, -wordSize));
1734 __ bind(skip);
1735
1736 // Align stack pointer for compiled code (note that caller is
1737 // responsible for undoing this fixup by remembering the old SP
1738 // in an rbp,-relative location)
1739 __ andptr(rsp, -(StackAlignmentInBytes));
1740
1741 // push the (possibly adjusted) return address
1742 __ push(rdi);
1743
1744 // and begin the OSR nmethod
1745 __ jmp(Address(rbx, nmethod::osr_entry_point_offset()));
1746 }
1747 }
1748 }
1749
1750
1751 void TemplateTable::if_0cmp(Condition cc) {
1752 transition(itos, vtos);
1753 // assume branch is more often taken than not (loops use backward branches)
1754 Label not_taken;
1755 __ testl(rax, rax);
1756 __ jcc(j_not(cc), not_taken);
1757 branch(false, false);
1758 __ bind(not_taken);
1759 __ profile_not_taken_branch(rax);
1760 }
1761
1762
1763 void TemplateTable::if_icmp(Condition cc) {
1764 transition(itos, vtos);
1765 // assume branch is more often taken than not (loops use backward branches)
1766 Label not_taken;
1767 __ pop_i(rdx);
1768 __ cmpl(rdx, rax);
1769 __ jcc(j_not(cc), not_taken);
1770 branch(false, false);
1771 __ bind(not_taken);
1772 __ profile_not_taken_branch(rax);
1773 }
1774
1775
1776 void TemplateTable::if_nullcmp(Condition cc) {
1777 transition(atos, vtos);
1778 // assume branch is more often taken than not (loops use backward branches)
1779 Label not_taken;
1780 __ testptr(rax, rax);
1781 __ jcc(j_not(cc), not_taken);
1782 branch(false, false);
1783 __ bind(not_taken);
1784 __ profile_not_taken_branch(rax);
1785 }
1786
1787
1788 void TemplateTable::if_acmp(Condition cc) {
1789 transition(atos, vtos);
1790 // assume branch is more often taken than not (loops use backward branches)
1791 Label not_taken;
1792 __ pop_ptr(rdx);
1793 __ cmpptr(rdx, rax);
1794 __ jcc(j_not(cc), not_taken);
1795 branch(false, false);
1796 __ bind(not_taken);
1797 __ profile_not_taken_branch(rax);
1798 }
1799
1800
1801 void TemplateTable::ret() {
1802 transition(vtos, vtos);
1803 locals_index(rbx);
1804 __ movptr(rbx, iaddress(rbx)); // get return bci, compute return bcp
1805 __ profile_ret(rbx, rcx);
1806 __ get_method(rax);
1807 __ movptr(rsi, Address(rax, methodOopDesc::const_offset()));
1808 __ lea(rsi, Address(rsi, rbx, Address::times_1,
1809 constMethodOopDesc::codes_offset()));
1810 __ dispatch_next(vtos);
1811 }
1812
1813
1814 void TemplateTable::wide_ret() {
1815 transition(vtos, vtos);
1816 locals_index_wide(rbx);
1817 __ movptr(rbx, iaddress(rbx)); // get return bci, compute return bcp
1818 __ profile_ret(rbx, rcx);
1819 __ get_method(rax);
1820 __ movptr(rsi, Address(rax, methodOopDesc::const_offset()));
1821 __ lea(rsi, Address(rsi, rbx, Address::times_1, constMethodOopDesc::codes_offset()));
1822 __ dispatch_next(vtos);
1823 }
1824
1825
1826 void TemplateTable::tableswitch() {
1827 Label default_case, continue_execution;
1828 transition(itos, vtos);
1829 // align rsi
1830 __ lea(rbx, at_bcp(wordSize));
1831 __ andptr(rbx, -wordSize);
1832 // load lo & hi
1833 __ movl(rcx, Address(rbx, 1 * wordSize));
1834 __ movl(rdx, Address(rbx, 2 * wordSize));
1835 __ bswapl(rcx);
1836 __ bswapl(rdx);
1837 // check against lo & hi
1838 __ cmpl(rax, rcx);
1839 __ jccb(Assembler::less, default_case);
1840 __ cmpl(rax, rdx);
1841 __ jccb(Assembler::greater, default_case);
1842 // lookup dispatch offset
1843 __ subl(rax, rcx);
1844 __ movl(rdx, Address(rbx, rax, Address::times_4, 3 * BytesPerInt));
1845 __ profile_switch_case(rax, rbx, rcx);
1846 // continue execution
1847 __ bind(continue_execution);
1848 __ bswapl(rdx);
1849 __ load_unsigned_byte(rbx, Address(rsi, rdx, Address::times_1));
1850 __ addptr(rsi, rdx);
1851 __ dispatch_only(vtos);
1852 // handle default
1853 __ bind(default_case);
1854 __ profile_switch_default(rax);
1855 __ movl(rdx, Address(rbx, 0));
1856 __ jmp(continue_execution);
1857 }
1858
1859
1860 void TemplateTable::lookupswitch() {
1861 transition(itos, itos);
1862 __ stop("lookupswitch bytecode should have been rewritten");
1863 }
1864
1865
1866 void TemplateTable::fast_linearswitch() {
1867 transition(itos, vtos);
1868 Label loop_entry, loop, found, continue_execution;
1869 // bswapl rax, so we can avoid bswapping the table entries
1870 __ bswapl(rax);
1871 // align rsi
1872 __ lea(rbx, at_bcp(wordSize)); // btw: should be able to get rid of this instruction (change offsets below)
1873 __ andptr(rbx, -wordSize);
1874 // set counter
1875 __ movl(rcx, Address(rbx, wordSize));
1876 __ bswapl(rcx);
1877 __ jmpb(loop_entry);
1878 // table search
1879 __ bind(loop);
1880 __ cmpl(rax, Address(rbx, rcx, Address::times_8, 2 * wordSize));
1881 __ jccb(Assembler::equal, found);
1882 __ bind(loop_entry);
1883 __ decrementl(rcx);
1884 __ jcc(Assembler::greaterEqual, loop);
1885 // default case
1886 __ profile_switch_default(rax);
1887 __ movl(rdx, Address(rbx, 0));
1888 __ jmpb(continue_execution);
1889 // entry found -> get offset
1890 __ bind(found);
1891 __ movl(rdx, Address(rbx, rcx, Address::times_8, 3 * wordSize));
1892 __ profile_switch_case(rcx, rax, rbx);
1893 // continue execution
1894 __ bind(continue_execution);
1895 __ bswapl(rdx);
1896 __ load_unsigned_byte(rbx, Address(rsi, rdx, Address::times_1));
1897 __ addptr(rsi, rdx);
1898 __ dispatch_only(vtos);
1899 }
1900
1901
1902 void TemplateTable::fast_binaryswitch() {
1903 transition(itos, vtos);
1904 // Implementation using the following core algorithm:
1905 //
1906 // int binary_search(int key, LookupswitchPair* array, int n) {
1907 // // Binary search according to "Methodik des Programmierens" by
1908 // // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985.
1909 // int i = 0;
1910 // int j = n;
1911 // while (i+1 < j) {
1912 // // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q)
1913 // // with Q: for all i: 0 <= i < n: key < a[i]
1914 // // where a stands for the array and assuming that the (inexisting)
1915 // // element a[n] is infinitely big.
1916 // int h = (i + j) >> 1;
1917 // // i < h < j
1918 // if (key < array[h].fast_match()) {
1919 // j = h;
1920 // } else {
1921 // i = h;
1922 // }
1923 // }
1924 // // R: a[i] <= key < a[i+1] or Q
1925 // // (i.e., if key is within array, i is the correct index)
1926 // return i;
1927 // }
1928
1929 // register allocation
1930 const Register key = rax; // already set (tosca)
1931 const Register array = rbx;
1932 const Register i = rcx;
1933 const Register j = rdx;
1934 const Register h = rdi; // needs to be restored
1935 const Register temp = rsi;
1936 // setup array
1937 __ save_bcp();
1938
1939 __ lea(array, at_bcp(3*wordSize)); // btw: should be able to get rid of this instruction (change offsets below)
1940 __ andptr(array, -wordSize);
1941 // initialize i & j
1942 __ xorl(i, i); // i = 0;
1943 __ movl(j, Address(array, -wordSize)); // j = length(array);
1944 // Convert j into native byteordering
1945 __ bswapl(j);
1946 // and start
1947 Label entry;
1948 __ jmp(entry);
1949
1950 // binary search loop
1951 { Label loop;
1952 __ bind(loop);
1953 // int h = (i + j) >> 1;
1954 __ leal(h, Address(i, j, Address::times_1)); // h = i + j;
1955 __ sarl(h, 1); // h = (i + j) >> 1;
1956 // if (key < array[h].fast_match()) {
1957 // j = h;
1958 // } else {
1959 // i = h;
1960 // }
1961 // Convert array[h].match to native byte-ordering before compare
1962 __ movl(temp, Address(array, h, Address::times_8, 0*wordSize));
1963 __ bswapl(temp);
1964 __ cmpl(key, temp);
1965 if (VM_Version::supports_cmov()) {
1966 __ cmovl(Assembler::less , j, h); // j = h if (key < array[h].fast_match())
1967 __ cmovl(Assembler::greaterEqual, i, h); // i = h if (key >= array[h].fast_match())
1968 } else {
1969 Label set_i, end_of_if;
1970 __ jccb(Assembler::greaterEqual, set_i); // {
1971 __ mov(j, h); // j = h;
1972 __ jmp(end_of_if); // }
1973 __ bind(set_i); // else {
1974 __ mov(i, h); // i = h;
1975 __ bind(end_of_if); // }
1976 }
1977 // while (i+1 < j)
1978 __ bind(entry);
1979 __ leal(h, Address(i, 1)); // i+1
1980 __ cmpl(h, j); // i+1 < j
1981 __ jcc(Assembler::less, loop);
1982 }
1983
1984 // end of binary search, result index is i (must check again!)
1985 Label default_case;
1986 // Convert array[i].match to native byte-ordering before compare
1987 __ movl(temp, Address(array, i, Address::times_8, 0*wordSize));
1988 __ bswapl(temp);
1989 __ cmpl(key, temp);
1990 __ jcc(Assembler::notEqual, default_case);
1991
1992 // entry found -> j = offset
1993 __ movl(j , Address(array, i, Address::times_8, 1*wordSize));
1994 __ profile_switch_case(i, key, array);
1995 __ bswapl(j);
1996 LP64_ONLY(__ movslq(j, j));
1997 __ restore_bcp();
1998 __ restore_locals(); // restore rdi
1999 __ load_unsigned_byte(rbx, Address(rsi, j, Address::times_1));
2000
2001 __ addptr(rsi, j);
2002 __ dispatch_only(vtos);
2003
2004 // default case -> j = default offset
2005 __ bind(default_case);
2006 __ profile_switch_default(i);
2007 __ movl(j, Address(array, -2*wordSize));
2008 __ bswapl(j);
2009 LP64_ONLY(__ movslq(j, j));
2010 __ restore_bcp();
2011 __ restore_locals(); // restore rdi
2012 __ load_unsigned_byte(rbx, Address(rsi, j, Address::times_1));
2013 __ addptr(rsi, j);
2014 __ dispatch_only(vtos);
2015 }
2016
2017
2018 void TemplateTable::_return(TosState state) {
2019 transition(state, state);
2020 assert(_desc->calls_vm(), "inconsistent calls_vm information"); // call in remove_activation
2021
2022 if (_desc->bytecode() == Bytecodes::_return_register_finalizer) {
2023 assert(state == vtos, "only valid state");
2024 __ movptr(rax, aaddress(0));
2025 __ movptr(rdi, Address(rax, oopDesc::klass_offset_in_bytes()));
2026 __ movl(rdi, Address(rdi, Klass::access_flags_offset_in_bytes() + sizeof(oopDesc)));
2027 __ testl(rdi, JVM_ACC_HAS_FINALIZER);
2028 Label skip_register_finalizer;
2029 __ jcc(Assembler::zero, skip_register_finalizer);
2030
2031 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), rax);
2032
2033 __ bind(skip_register_finalizer);
2034 }
2035
2036 __ remove_activation(state, rsi);
2037 __ jmp(rsi);
2038 }
2039
2040
2041 // ----------------------------------------------------------------------------
2042 // Volatile variables demand their effects be made known to all CPU's in
2043 // order. Store buffers on most chips allow reads & writes to reorder; the
2044 // JMM's ReadAfterWrite.java test fails in -Xint mode without some kind of
2045 // memory barrier (i.e., it's not sufficient that the interpreter does not
2046 // reorder volatile references, the hardware also must not reorder them).
2047 //
2048 // According to the new Java Memory Model (JMM):
2049 // (1) All volatiles are serialized wrt to each other.
2050 // ALSO reads & writes act as aquire & release, so:
2051 // (2) A read cannot let unrelated NON-volatile memory refs that happen after
2052 // the read float up to before the read. It's OK for non-volatile memory refs
2053 // that happen before the volatile read to float down below it.
2054 // (3) Similar a volatile write cannot let unrelated NON-volatile memory refs
2055 // that happen BEFORE the write float down to after the write. It's OK for
2056 // non-volatile memory refs that happen after the volatile write to float up
2057 // before it.
2058 //
2059 // We only put in barriers around volatile refs (they are expensive), not
2060 // _between_ memory refs (that would require us to track the flavor of the
2061 // previous memory refs). Requirements (2) and (3) require some barriers
2062 // before volatile stores and after volatile loads. These nearly cover
2063 // requirement (1) but miss the volatile-store-volatile-load case. This final
2064 // case is placed after volatile-stores although it could just as well go
2065 // before volatile-loads.
2066 void TemplateTable::volatile_barrier(Assembler::Membar_mask_bits order_constraint ) {
2067 // Helper function to insert a is-volatile test and memory barrier
2068 if( !os::is_MP() ) return; // Not needed on single CPU
2069 __ membar(order_constraint);
2070 }
2071
2072 void TemplateTable::resolve_cache_and_index(int byte_no,
2073 Register result,
2074 Register Rcache,
2075 Register index,
2076 size_t index_size) {
2077 Register temp = rbx;
2078
2079 assert_different_registers(result, Rcache, index, temp);
2080
2081 Label resolved;
2082 __ get_cache_and_index_at_bcp(Rcache, index, 1, index_size);
2083 if (byte_no == f1_oop) {
2084 // We are resolved if the f1 field contains a non-null object (CallSite, etc.)
2085 // This kind of CP cache entry does not need to match the flags byte, because
2086 // there is a 1-1 relation between bytecode type and CP entry type.
2087 assert(result != noreg, ""); //else do cmpptr(Address(...), (int32_t) NULL_WORD)
2088 __ movptr(result, Address(Rcache, index, Address::times_ptr, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::f1_offset()));
2089 __ testptr(result, result);
2090 __ jcc(Assembler::notEqual, resolved);
2091 } else {
2092 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
2093 assert(result == noreg, ""); //else change code for setting result
2094 const int shift_count = (1 + byte_no)*BitsPerByte;
2095 __ movl(temp, Address(Rcache, index, Address::times_4, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::indices_offset()));
2096 __ shrl(temp, shift_count);
2097 // have we resolved this bytecode?
2098 __ andl(temp, 0xFF);
2099 __ cmpl(temp, (int)bytecode());
2100 __ jcc(Assembler::equal, resolved);
2101 }
2102
2103 // resolve first time through
2104 address entry;
2105 switch (bytecode()) {
2106 case Bytecodes::_getstatic : // fall through
2107 case Bytecodes::_putstatic : // fall through
2108 case Bytecodes::_getfield : // fall through
2109 case Bytecodes::_putfield : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_get_put); break;
2110 case Bytecodes::_invokevirtual : // fall through
2111 case Bytecodes::_invokespecial : // fall through
2112 case Bytecodes::_invokestatic : // fall through
2113 case Bytecodes::_invokeinterface: entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invoke); break;
2114 case Bytecodes::_invokedynamic : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invokedynamic); break;
2115 case Bytecodes::_fast_aldc : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc); break;
2116 case Bytecodes::_fast_aldc_w : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc); break;
2117 default : ShouldNotReachHere(); break;
2118 }
2119 __ movl(temp, (int)bytecode());
2120 __ call_VM(noreg, entry, temp);
2121 // Update registers with resolved info
2122 __ get_cache_and_index_at_bcp(Rcache, index, 1, index_size);
2123 if (result != noreg)
2124 __ movptr(result, Address(Rcache, index, Address::times_ptr, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::f1_offset()));
2125 __ bind(resolved);
2126 }
2127
2128
2129 // The cache and index registers must be set before call
2130 void TemplateTable::load_field_cp_cache_entry(Register obj,
2131 Register cache,
2132 Register index,
2133 Register off,
2134 Register flags,
2135 bool is_static = false) {
2136 assert_different_registers(cache, index, flags, off);
2137
2138 ByteSize cp_base_offset = constantPoolCacheOopDesc::base_offset();
2139 // Field offset
2140 __ movptr(off, Address(cache, index, Address::times_ptr,
2141 in_bytes(cp_base_offset + ConstantPoolCacheEntry::f2_offset())));
2142 // Flags
2143 __ movl(flags, Address(cache, index, Address::times_ptr,
2144 in_bytes(cp_base_offset + ConstantPoolCacheEntry::flags_offset())));
2145
2146 // klass overwrite register
2147 if (is_static) {
2148 __ movptr(obj, Address(cache, index, Address::times_ptr,
2149 in_bytes(cp_base_offset + ConstantPoolCacheEntry::f1_offset())));
2150 }
2151 }
2152
2153 void TemplateTable::load_invoke_cp_cache_entry(int byte_no,
2154 Register method,
2155 Register itable_index,
2156 Register flags,
2157 bool is_invokevirtual,
2158 bool is_invokevfinal /*unused*/,
2159 bool is_invokedynamic) {
2160 // setup registers
2161 const Register cache = rcx;
2162 const Register index = rdx;
2163 assert_different_registers(method, flags);
2164 assert_different_registers(method, cache, index);
2165 assert_different_registers(itable_index, flags);
2166 assert_different_registers(itable_index, cache, index);
2167 // determine constant pool cache field offsets
2168 const int method_offset = in_bytes(
2169 constantPoolCacheOopDesc::base_offset() +
2170 (is_invokevirtual
2171 ? ConstantPoolCacheEntry::f2_offset()
2172 : ConstantPoolCacheEntry::f1_offset()
2173 )
2174 );
2175 const int flags_offset = in_bytes(constantPoolCacheOopDesc::base_offset() +
2176 ConstantPoolCacheEntry::flags_offset());
2177 // access constant pool cache fields
2178 const int index_offset = in_bytes(constantPoolCacheOopDesc::base_offset() +
2179 ConstantPoolCacheEntry::f2_offset());
2180
2181 if (byte_no == f1_oop) {
2182 // Resolved f1_oop goes directly into 'method' register.
2183 assert(is_invokedynamic, "");
2184 resolve_cache_and_index(byte_no, method, cache, index, sizeof(u4));
2185 } else {
2186 resolve_cache_and_index(byte_no, noreg, cache, index, sizeof(u2));
2187 __ movptr(method, Address(cache, index, Address::times_ptr, method_offset));
2188 }
2189 if (itable_index != noreg) {
2190 __ movptr(itable_index, Address(cache, index, Address::times_ptr, index_offset));
2191 }
2192 __ movl(flags, Address(cache, index, Address::times_ptr, flags_offset));
2193 }
2194
2195
2196 // The registers cache and index expected to be set before call.
2197 // Correct values of the cache and index registers are preserved.
2198 void TemplateTable::jvmti_post_field_access(Register cache,
2199 Register index,
2200 bool is_static,
2201 bool has_tos) {
2202 if (JvmtiExport::can_post_field_access()) {
2203 // Check to see if a field access watch has been set before we take
2204 // the time to call into the VM.
2205 Label L1;
2206 assert_different_registers(cache, index, rax);
2207 __ mov32(rax, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
2208 __ testl(rax,rax);
2209 __ jcc(Assembler::zero, L1);
2210
2211 // cache entry pointer
2212 __ addptr(cache, in_bytes(constantPoolCacheOopDesc::base_offset()));
2213 __ shll(index, LogBytesPerWord);
2214 __ addptr(cache, index);
2215 if (is_static) {
2216 __ xorptr(rax, rax); // NULL object reference
2217 } else {
2218 __ pop(atos); // Get the object
2219 __ verify_oop(rax);
2220 __ push(atos); // Restore stack state
2221 }
2222 // rax,: object pointer or NULL
2223 // cache: cache entry pointer
2224 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access),
2225 rax, cache);
2226 __ get_cache_and_index_at_bcp(cache, index, 1);
2227 __ bind(L1);
2228 }
2229 }
2230
2231 void TemplateTable::pop_and_check_object(Register r) {
2232 __ pop_ptr(r);
2233 __ null_check(r); // for field access must check obj.
2234 __ verify_oop(r);
2235 }
2236
2237 void TemplateTable::getfield_or_static(int byte_no, bool is_static) {
2238 transition(vtos, vtos);
2239
2240 const Register cache = rcx;
2241 const Register index = rdx;
2242 const Register obj = rcx;
2243 const Register off = rbx;
2244 const Register flags = rax;
2245
2246 resolve_cache_and_index(byte_no, noreg, cache, index, sizeof(u2));
2247 jvmti_post_field_access(cache, index, is_static, false);
2248 load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
2249
2250 if (!is_static) pop_and_check_object(obj);
2251
2252 const Address lo(obj, off, Address::times_1, 0*wordSize);
2253 const Address hi(obj, off, Address::times_1, 1*wordSize);
2254
2255 Label Done, notByte, notInt, notShort, notChar, notLong, notFloat, notObj, notDouble;
2256
2257 __ shrl(flags, ConstantPoolCacheEntry::tosBits);
2258 assert(btos == 0, "change code, btos != 0");
2259 // btos
2260 __ andptr(flags, 0x0f);
2261 __ jcc(Assembler::notZero, notByte);
2262
2263 __ load_signed_byte(rax, lo );
2264 __ push(btos);
2265 // Rewrite bytecode to be faster
2266 if (!is_static) {
2267 patch_bytecode(Bytecodes::_fast_bgetfield, rcx, rbx);
2268 }
2269 __ jmp(Done);
2270
2271 __ bind(notByte);
2272 // itos
2273 __ cmpl(flags, itos );
2274 __ jcc(Assembler::notEqual, notInt);
2275
2276 __ movl(rax, lo );
2277 __ push(itos);
2278 // Rewrite bytecode to be faster
2279 if (!is_static) {
2280 patch_bytecode(Bytecodes::_fast_igetfield, rcx, rbx);
2281 }
2282 __ jmp(Done);
2283
2284 __ bind(notInt);
2285 // atos
2286 __ cmpl(flags, atos );
2287 __ jcc(Assembler::notEqual, notObj);
2288
2289 __ movl(rax, lo );
2290 __ push(atos);
2291 if (!is_static) {
2292 patch_bytecode(Bytecodes::_fast_agetfield, rcx, rbx);
2293 }
2294 __ jmp(Done);
2295
2296 __ bind(notObj);
2297 // ctos
2298 __ cmpl(flags, ctos );
2299 __ jcc(Assembler::notEqual, notChar);
2300
2301 __ load_unsigned_short(rax, lo );
2302 __ push(ctos);
2303 if (!is_static) {
2304 patch_bytecode(Bytecodes::_fast_cgetfield, rcx, rbx);
2305 }
2306 __ jmp(Done);
2307
2308 __ bind(notChar);
2309 // stos
2310 __ cmpl(flags, stos );
2311 __ jcc(Assembler::notEqual, notShort);
2312
2313 __ load_signed_short(rax, lo );
2314 __ push(stos);
2315 if (!is_static) {
2316 patch_bytecode(Bytecodes::_fast_sgetfield, rcx, rbx);
2317 }
2318 __ jmp(Done);
2319
2320 __ bind(notShort);
2321 // ltos
2322 __ cmpl(flags, ltos );
2323 __ jcc(Assembler::notEqual, notLong);
2324
2325 // Generate code as if volatile. There just aren't enough registers to
2326 // save that information and this code is faster than the test.
2327 __ fild_d(lo); // Must load atomically
2328 __ subptr(rsp,2*wordSize); // Make space for store
2329 __ fistp_d(Address(rsp,0));
2330 __ pop(rax);
2331 __ pop(rdx);
2332
2333 __ push(ltos);
2334 // Don't rewrite to _fast_lgetfield for potential volatile case.
2335 __ jmp(Done);
2336
2337 __ bind(notLong);
2338 // ftos
2339 __ cmpl(flags, ftos );
2340 __ jcc(Assembler::notEqual, notFloat);
2341
2342 __ fld_s(lo);
2343 __ push(ftos);
2344 if (!is_static) {
2345 patch_bytecode(Bytecodes::_fast_fgetfield, rcx, rbx);
2346 }
2347 __ jmp(Done);
2348
2349 __ bind(notFloat);
2350 // dtos
2351 __ cmpl(flags, dtos );
2352 __ jcc(Assembler::notEqual, notDouble);
2353
2354 __ fld_d(lo);
2355 __ push(dtos);
2356 if (!is_static) {
2357 patch_bytecode(Bytecodes::_fast_dgetfield, rcx, rbx);
2358 }
2359 __ jmpb(Done);
2360
2361 __ bind(notDouble);
2362
2363 __ stop("Bad state");
2364
2365 __ bind(Done);
2366 // Doug Lea believes this is not needed with current Sparcs (TSO) and Intel (PSO).
2367 // volatile_barrier( );
2368 }
2369
2370
2371 void TemplateTable::getfield(int byte_no) {
2372 getfield_or_static(byte_no, false);
2373 }
2374
2375
2376 void TemplateTable::getstatic(int byte_no) {
2377 getfield_or_static(byte_no, true);
2378 }
2379
2380 // The registers cache and index expected to be set before call.
2381 // The function may destroy various registers, just not the cache and index registers.
2382 void TemplateTable::jvmti_post_field_mod(Register cache, Register index, bool is_static) {
2383
2384 ByteSize cp_base_offset = constantPoolCacheOopDesc::base_offset();
2385
2386 if (JvmtiExport::can_post_field_modification()) {
2387 // Check to see if a field modification watch has been set before we take
2388 // the time to call into the VM.
2389 Label L1;
2390 assert_different_registers(cache, index, rax);
2391 __ mov32(rax, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
2392 __ testl(rax, rax);
2393 __ jcc(Assembler::zero, L1);
2394
2395 // The cache and index registers have been already set.
2396 // This allows to eliminate this call but the cache and index
2397 // registers have to be correspondingly used after this line.
2398 __ get_cache_and_index_at_bcp(rax, rdx, 1);
2399
2400 if (is_static) {
2401 // Life is simple. Null out the object pointer.
2402 __ xorptr(rbx, rbx);
2403 } else {
2404 // Life is harder. The stack holds the value on top, followed by the object.
2405 // We don't know the size of the value, though; it could be one or two words
2406 // depending on its type. As a result, we must find the type to determine where
2407 // the object is.
2408 Label two_word, valsize_known;
2409 __ movl(rcx, Address(rax, rdx, Address::times_ptr, in_bytes(cp_base_offset +
2410 ConstantPoolCacheEntry::flags_offset())));
2411 __ mov(rbx, rsp);
2412 __ shrl(rcx, ConstantPoolCacheEntry::tosBits);
2413 // Make sure we don't need to mask rcx for tosBits after the above shift
2414 ConstantPoolCacheEntry::verify_tosBits();
2415 __ cmpl(rcx, ltos);
2416 __ jccb(Assembler::equal, two_word);
2417 __ cmpl(rcx, dtos);
2418 __ jccb(Assembler::equal, two_word);
2419 __ addptr(rbx, Interpreter::expr_offset_in_bytes(1)); // one word jvalue (not ltos, dtos)
2420 __ jmpb(valsize_known);
2421
2422 __ bind(two_word);
2423 __ addptr(rbx, Interpreter::expr_offset_in_bytes(2)); // two words jvalue
2424
2425 __ bind(valsize_known);
2426 // setup object pointer
2427 __ movptr(rbx, Address(rbx, 0));
2428 }
2429 // cache entry pointer
2430 __ addptr(rax, in_bytes(cp_base_offset));
2431 __ shll(rdx, LogBytesPerWord);
2432 __ addptr(rax, rdx);
2433 // object (tos)
2434 __ mov(rcx, rsp);
2435 // rbx,: object pointer set up above (NULL if static)
2436 // rax,: cache entry pointer
2437 // rcx: jvalue object on the stack
2438 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification),
2439 rbx, rax, rcx);
2440 __ get_cache_and_index_at_bcp(cache, index, 1);
2441 __ bind(L1);
2442 }
2443 }
2444
2445
2446 void TemplateTable::putfield_or_static(int byte_no, bool is_static) {
2447 transition(vtos, vtos);
2448
2449 const Register cache = rcx;
2450 const Register index = rdx;
2451 const Register obj = rcx;
2452 const Register off = rbx;
2453 const Register flags = rax;
2454
2455 resolve_cache_and_index(byte_no, noreg, cache, index, sizeof(u2));
2456 jvmti_post_field_mod(cache, index, is_static);
2457 load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
2458
2459 // Doug Lea believes this is not needed with current Sparcs (TSO) and Intel (PSO).
2460 // volatile_barrier( );
2461
2462 Label notVolatile, Done;
2463 __ movl(rdx, flags);
2464 __ shrl(rdx, ConstantPoolCacheEntry::volatileField);
2465 __ andl(rdx, 0x1);
2466
2467 // field addresses
2468 const Address lo(obj, off, Address::times_1, 0*wordSize);
2469 const Address hi(obj, off, Address::times_1, 1*wordSize);
2470
2471 Label notByte, notInt, notShort, notChar, notLong, notFloat, notObj, notDouble;
2472
2473 __ shrl(flags, ConstantPoolCacheEntry::tosBits);
2474 assert(btos == 0, "change code, btos != 0");
2475 // btos
2476 __ andl(flags, 0x0f);
2477 __ jcc(Assembler::notZero, notByte);
2478
2479 __ pop(btos);
2480 if (!is_static) pop_and_check_object(obj);
2481 __ movb(lo, rax );
2482 if (!is_static) {
2483 patch_bytecode(Bytecodes::_fast_bputfield, rcx, rbx);
2484 }
2485 __ jmp(Done);
2486
2487 __ bind(notByte);
2488 // itos
2489 __ cmpl(flags, itos );
2490 __ jcc(Assembler::notEqual, notInt);
2491
2492 __ pop(itos);
2493 if (!is_static) pop_and_check_object(obj);
2494
2495 __ movl(lo, rax );
2496 if (!is_static) {
2497 patch_bytecode(Bytecodes::_fast_iputfield, rcx, rbx);
2498 }
2499 __ jmp(Done);
2500
2501 __ bind(notInt);
2502 // atos
2503 __ cmpl(flags, atos );
2504 __ jcc(Assembler::notEqual, notObj);
2505
2506 __ pop(atos);
2507 if (!is_static) pop_and_check_object(obj);
2508
2509 do_oop_store(_masm, lo, rax, _bs->kind(), false);
2510
2511 if (!is_static) {
2512 patch_bytecode(Bytecodes::_fast_aputfield, rcx, rbx);
2513 }
2514
2515 __ jmp(Done);
2516
2517 __ bind(notObj);
2518 // ctos
2519 __ cmpl(flags, ctos );
2520 __ jcc(Assembler::notEqual, notChar);
2521
2522 __ pop(ctos);
2523 if (!is_static) pop_and_check_object(obj);
2524 __ movw(lo, rax );
2525 if (!is_static) {
2526 patch_bytecode(Bytecodes::_fast_cputfield, rcx, rbx);
2527 }
2528 __ jmp(Done);
2529
2530 __ bind(notChar);
2531 // stos
2532 __ cmpl(flags, stos );
2533 __ jcc(Assembler::notEqual, notShort);
2534
2535 __ pop(stos);
2536 if (!is_static) pop_and_check_object(obj);
2537 __ movw(lo, rax );
2538 if (!is_static) {
2539 patch_bytecode(Bytecodes::_fast_sputfield, rcx, rbx);
2540 }
2541 __ jmp(Done);
2542
2543 __ bind(notShort);
2544 // ltos
2545 __ cmpl(flags, ltos );
2546 __ jcc(Assembler::notEqual, notLong);
2547
2548 Label notVolatileLong;
2549 __ testl(rdx, rdx);
2550 __ jcc(Assembler::zero, notVolatileLong);
2551
2552 __ pop(ltos); // overwrites rdx, do this after testing volatile.
2553 if (!is_static) pop_and_check_object(obj);
2554
2555 // Replace with real volatile test
2556 __ push(rdx);
2557 __ push(rax); // Must update atomically with FIST
2558 __ fild_d(Address(rsp,0)); // So load into FPU register
2559 __ fistp_d(lo); // and put into memory atomically
2560 __ addptr(rsp, 2*wordSize);
2561 // volatile_barrier();
2562 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
2563 Assembler::StoreStore));
2564 // Don't rewrite volatile version
2565 __ jmp(notVolatile);
2566
2567 __ bind(notVolatileLong);
2568
2569 __ pop(ltos); // overwrites rdx
2570 if (!is_static) pop_and_check_object(obj);
2571 NOT_LP64(__ movptr(hi, rdx));
2572 __ movptr(lo, rax);
2573 if (!is_static) {
2574 patch_bytecode(Bytecodes::_fast_lputfield, rcx, rbx);
2575 }
2576 __ jmp(notVolatile);
2577
2578 __ bind(notLong);
2579 // ftos
2580 __ cmpl(flags, ftos );
2581 __ jcc(Assembler::notEqual, notFloat);
2582
2583 __ pop(ftos);
2584 if (!is_static) pop_and_check_object(obj);
2585 __ fstp_s(lo);
2586 if (!is_static) {
2587 patch_bytecode(Bytecodes::_fast_fputfield, rcx, rbx);
2588 }
2589 __ jmp(Done);
2590
2591 __ bind(notFloat);
2592 // dtos
2593 __ cmpl(flags, dtos );
2594 __ jcc(Assembler::notEqual, notDouble);
2595
2596 __ pop(dtos);
2597 if (!is_static) pop_and_check_object(obj);
2598 __ fstp_d(lo);
2599 if (!is_static) {
2600 patch_bytecode(Bytecodes::_fast_dputfield, rcx, rbx);
2601 }
2602 __ jmp(Done);
2603
2604 __ bind(notDouble);
2605
2606 __ stop("Bad state");
2607
2608 __ bind(Done);
2609
2610 // Check for volatile store
2611 __ testl(rdx, rdx);
2612 __ jcc(Assembler::zero, notVolatile);
2613 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
2614 Assembler::StoreStore));
2615 __ bind(notVolatile);
2616 }
2617
2618
2619 void TemplateTable::putfield(int byte_no) {
2620 putfield_or_static(byte_no, false);
2621 }
2622
2623
2624 void TemplateTable::putstatic(int byte_no) {
2625 putfield_or_static(byte_no, true);
2626 }
2627
2628 void TemplateTable::jvmti_post_fast_field_mod() {
2629 if (JvmtiExport::can_post_field_modification()) {
2630 // Check to see if a field modification watch has been set before we take
2631 // the time to call into the VM.
2632 Label L2;
2633 __ mov32(rcx, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
2634 __ testl(rcx,rcx);
2635 __ jcc(Assembler::zero, L2);
2636 __ pop_ptr(rbx); // copy the object pointer from tos
2637 __ verify_oop(rbx);
2638 __ push_ptr(rbx); // put the object pointer back on tos
2639 __ subptr(rsp, sizeof(jvalue)); // add space for a jvalue object
2640 __ mov(rcx, rsp);
2641 __ push_ptr(rbx); // save object pointer so we can steal rbx,
2642 __ xorptr(rbx, rbx);
2643 const Address lo_value(rcx, rbx, Address::times_1, 0*wordSize);
2644 const Address hi_value(rcx, rbx, Address::times_1, 1*wordSize);
2645 switch (bytecode()) { // load values into the jvalue object
2646 case Bytecodes::_fast_bputfield: __ movb(lo_value, rax); break;
2647 case Bytecodes::_fast_sputfield: __ movw(lo_value, rax); break;
2648 case Bytecodes::_fast_cputfield: __ movw(lo_value, rax); break;
2649 case Bytecodes::_fast_iputfield: __ movl(lo_value, rax); break;
2650 case Bytecodes::_fast_lputfield:
2651 NOT_LP64(__ movptr(hi_value, rdx));
2652 __ movptr(lo_value, rax);
2653 break;
2654
2655 // need to call fld_s() after fstp_s() to restore the value for below
2656 case Bytecodes::_fast_fputfield: __ fstp_s(lo_value); __ fld_s(lo_value); break;
2657
2658 // need to call fld_d() after fstp_d() to restore the value for below
2659 case Bytecodes::_fast_dputfield: __ fstp_d(lo_value); __ fld_d(lo_value); break;
2660
2661 // since rcx is not an object we don't call store_check() here
2662 case Bytecodes::_fast_aputfield: __ movptr(lo_value, rax); break;
2663
2664 default: ShouldNotReachHere();
2665 }
2666 __ pop_ptr(rbx); // restore copy of object pointer
2667
2668 // Save rax, and sometimes rdx because call_VM() will clobber them,
2669 // then use them for JVM/DI purposes
2670 __ push(rax);
2671 if (bytecode() == Bytecodes::_fast_lputfield) __ push(rdx);
2672 // access constant pool cache entry
2673 __ get_cache_entry_pointer_at_bcp(rax, rdx, 1);
2674 __ verify_oop(rbx);
2675 // rbx,: object pointer copied above
2676 // rax,: cache entry pointer
2677 // rcx: jvalue object on the stack
2678 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), rbx, rax, rcx);
2679 if (bytecode() == Bytecodes::_fast_lputfield) __ pop(rdx); // restore high value
2680 __ pop(rax); // restore lower value
2681 __ addptr(rsp, sizeof(jvalue)); // release jvalue object space
2682 __ bind(L2);
2683 }
2684 }
2685
2686 void TemplateTable::fast_storefield(TosState state) {
2687 transition(state, vtos);
2688
2689 ByteSize base = constantPoolCacheOopDesc::base_offset();
2690
2691 jvmti_post_fast_field_mod();
2692
2693 // access constant pool cache
2694 __ get_cache_and_index_at_bcp(rcx, rbx, 1);
2695
2696 // test for volatile with rdx but rdx is tos register for lputfield.
2697 if (bytecode() == Bytecodes::_fast_lputfield) __ push(rdx);
2698 __ movl(rdx, Address(rcx, rbx, Address::times_ptr, in_bytes(base +
2699 ConstantPoolCacheEntry::flags_offset())));
2700
2701 // replace index with field offset from cache entry
2702 __ movptr(rbx, Address(rcx, rbx, Address::times_ptr, in_bytes(base + ConstantPoolCacheEntry::f2_offset())));
2703
2704 // Doug Lea believes this is not needed with current Sparcs (TSO) and Intel (PSO).
2705 // volatile_barrier( );
2706
2707 Label notVolatile, Done;
2708 __ shrl(rdx, ConstantPoolCacheEntry::volatileField);
2709 __ andl(rdx, 0x1);
2710 // Check for volatile store
2711 __ testl(rdx, rdx);
2712 __ jcc(Assembler::zero, notVolatile);
2713
2714 if (bytecode() == Bytecodes::_fast_lputfield) __ pop(rdx);
2715
2716 // Get object from stack
2717 pop_and_check_object(rcx);
2718
2719 // field addresses
2720 const Address lo(rcx, rbx, Address::times_1, 0*wordSize);
2721 const Address hi(rcx, rbx, Address::times_1, 1*wordSize);
2722
2723 // access field
2724 switch (bytecode()) {
2725 case Bytecodes::_fast_bputfield: __ movb(lo, rax); break;
2726 case Bytecodes::_fast_sputfield: // fall through
2727 case Bytecodes::_fast_cputfield: __ movw(lo, rax); break;
2728 case Bytecodes::_fast_iputfield: __ movl(lo, rax); break;
2729 case Bytecodes::_fast_lputfield:
2730 NOT_LP64(__ movptr(hi, rdx));
2731 __ movptr(lo, rax);
2732 break;
2733 case Bytecodes::_fast_fputfield: __ fstp_s(lo); break;
2734 case Bytecodes::_fast_dputfield: __ fstp_d(lo); break;
2735 case Bytecodes::_fast_aputfield: {
2736 do_oop_store(_masm, lo, rax, _bs->kind(), false);
2737 break;
2738 }
2739 default:
2740 ShouldNotReachHere();
2741 }
2742
2743 Label done;
2744 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
2745 Assembler::StoreStore));
2746 // Barriers are so large that short branch doesn't reach!
2747 __ jmp(done);
2748
2749 // Same code as above, but don't need rdx to test for volatile.
2750 __ bind(notVolatile);
2751
2752 if (bytecode() == Bytecodes::_fast_lputfield) __ pop(rdx);
2753
2754 // Get object from stack
2755 pop_and_check_object(rcx);
2756
2757 // access field
2758 switch (bytecode()) {
2759 case Bytecodes::_fast_bputfield: __ movb(lo, rax); break;
2760 case Bytecodes::_fast_sputfield: // fall through
2761 case Bytecodes::_fast_cputfield: __ movw(lo, rax); break;
2762 case Bytecodes::_fast_iputfield: __ movl(lo, rax); break;
2763 case Bytecodes::_fast_lputfield:
2764 NOT_LP64(__ movptr(hi, rdx));
2765 __ movptr(lo, rax);
2766 break;
2767 case Bytecodes::_fast_fputfield: __ fstp_s(lo); break;
2768 case Bytecodes::_fast_dputfield: __ fstp_d(lo); break;
2769 case Bytecodes::_fast_aputfield: {
2770 do_oop_store(_masm, lo, rax, _bs->kind(), false);
2771 break;
2772 }
2773 default:
2774 ShouldNotReachHere();
2775 }
2776 __ bind(done);
2777 }
2778
2779
2780 void TemplateTable::fast_accessfield(TosState state) {
2781 transition(atos, state);
2782
2783 // do the JVMTI work here to avoid disturbing the register state below
2784 if (JvmtiExport::can_post_field_access()) {
2785 // Check to see if a field access watch has been set before we take
2786 // the time to call into the VM.
2787 Label L1;
2788 __ mov32(rcx, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
2789 __ testl(rcx,rcx);
2790 __ jcc(Assembler::zero, L1);
2791 // access constant pool cache entry
2792 __ get_cache_entry_pointer_at_bcp(rcx, rdx, 1);
2793 __ push_ptr(rax); // save object pointer before call_VM() clobbers it
2794 __ verify_oop(rax);
2795 // rax,: object pointer copied above
2796 // rcx: cache entry pointer
2797 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), rax, rcx);
2798 __ pop_ptr(rax); // restore object pointer
2799 __ bind(L1);
2800 }
2801
2802 // access constant pool cache
2803 __ get_cache_and_index_at_bcp(rcx, rbx, 1);
2804 // replace index with field offset from cache entry
2805 __ movptr(rbx, Address(rcx,
2806 rbx,
2807 Address::times_ptr,
2808 in_bytes(constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::f2_offset())));
2809
2810
2811 // rax,: object
2812 __ verify_oop(rax);
2813 __ null_check(rax);
2814 // field addresses
2815 const Address lo = Address(rax, rbx, Address::times_1, 0*wordSize);
2816 const Address hi = Address(rax, rbx, Address::times_1, 1*wordSize);
2817
2818 // access field
2819 switch (bytecode()) {
2820 case Bytecodes::_fast_bgetfield: __ movsbl(rax, lo ); break;
2821 case Bytecodes::_fast_sgetfield: __ load_signed_short(rax, lo ); break;
2822 case Bytecodes::_fast_cgetfield: __ load_unsigned_short(rax, lo ); break;
2823 case Bytecodes::_fast_igetfield: __ movl(rax, lo); break;
2824 case Bytecodes::_fast_lgetfield: __ stop("should not be rewritten"); break;
2825 case Bytecodes::_fast_fgetfield: __ fld_s(lo); break;
2826 case Bytecodes::_fast_dgetfield: __ fld_d(lo); break;
2827 case Bytecodes::_fast_agetfield: __ movptr(rax, lo); __ verify_oop(rax); break;
2828 default:
2829 ShouldNotReachHere();
2830 }
2831
2832 // Doug Lea believes this is not needed with current Sparcs(TSO) and Intel(PSO)
2833 // volatile_barrier( );
2834 }
2835
2836 void TemplateTable::fast_xaccess(TosState state) {
2837 transition(vtos, state);
2838 // get receiver
2839 __ movptr(rax, aaddress(0));
2840 // access constant pool cache
2841 __ get_cache_and_index_at_bcp(rcx, rdx, 2);
2842 __ movptr(rbx, Address(rcx,
2843 rdx,
2844 Address::times_ptr,
2845 in_bytes(constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::f2_offset())));
2846 // make sure exception is reported in correct bcp range (getfield is next instruction)
2847 __ increment(rsi);
2848 __ null_check(rax);
2849 const Address lo = Address(rax, rbx, Address::times_1, 0*wordSize);
2850 if (state == itos) {
2851 __ movl(rax, lo);
2852 } else if (state == atos) {
2853 __ movptr(rax, lo);
2854 __ verify_oop(rax);
2855 } else if (state == ftos) {
2856 __ fld_s(lo);
2857 } else {
2858 ShouldNotReachHere();
2859 }
2860 __ decrement(rsi);
2861 }
2862
2863
2864
2865 //----------------------------------------------------------------------------------------------------
2866 // Calls
2867
2868 void TemplateTable::count_calls(Register method, Register temp) {
2869 // implemented elsewhere
2870 ShouldNotReachHere();
2871 }
2872
2873
2874 void TemplateTable::prepare_invoke(Register method, Register index, int byte_no) {
2875 // determine flags
2876 Bytecodes::Code code = bytecode();
2877 const bool is_invokeinterface = code == Bytecodes::_invokeinterface;
2878 const bool is_invokedynamic = code == Bytecodes::_invokedynamic;
2879 const bool is_invokevirtual = code == Bytecodes::_invokevirtual;
2880 const bool is_invokespecial = code == Bytecodes::_invokespecial;
2881 const bool load_receiver = (code != Bytecodes::_invokestatic && code != Bytecodes::_invokedynamic);
2882 const bool receiver_null_check = is_invokespecial;
2883 const bool save_flags = is_invokeinterface || is_invokevirtual;
2884 // setup registers & access constant pool cache
2885 const Register recv = rcx;
2886 const Register flags = rdx;
2887 assert_different_registers(method, index, recv, flags);
2888
2889 // save 'interpreter return address'
2890 __ save_bcp();
2891
2892 load_invoke_cp_cache_entry(byte_no, method, index, flags, is_invokevirtual, false, is_invokedynamic);
2893
2894 // load receiver if needed (note: no return address pushed yet)
2895 if (load_receiver) {
2896 assert(!is_invokedynamic, "");
2897 __ movl(recv, flags);
2898 __ andl(recv, 0xFF);
2899 // recv count is 0 based?
2900 Address recv_addr(rsp, recv, Interpreter::stackElementScale(), -Interpreter::expr_offset_in_bytes(1));
2901 __ movptr(recv, recv_addr);
2902 __ verify_oop(recv);
2903 }
2904
2905 // do null check if needed
2906 if (receiver_null_check) {
2907 __ null_check(recv);
2908 }
2909
2910 if (save_flags) {
2911 __ mov(rsi, flags);
2912 }
2913
2914 // compute return type
2915 __ shrl(flags, ConstantPoolCacheEntry::tosBits);
2916 // Make sure we don't need to mask flags for tosBits after the above shift
2917 ConstantPoolCacheEntry::verify_tosBits();
2918 // load return address
2919 {
2920 address table_addr;
2921 if (is_invokeinterface || is_invokedynamic)
2922 table_addr = (address)Interpreter::return_5_addrs_by_index_table();
2923 else
2924 table_addr = (address)Interpreter::return_3_addrs_by_index_table();
2925 ExternalAddress table(table_addr);
2926 __ movptr(flags, ArrayAddress(table, Address(noreg, flags, Address::times_ptr)));
2927 }
2928
2929 // push return address
2930 __ push(flags);
2931
2932 // Restore flag value from the constant pool cache, and restore rsi
2933 // for later null checks. rsi is the bytecode pointer
2934 if (save_flags) {
2935 __ mov(flags, rsi);
2936 __ restore_bcp();
2937 }
2938 }
2939
2940
2941 void TemplateTable::invokevirtual_helper(Register index, Register recv,
2942 Register flags) {
2943
2944 // Uses temporary registers rax, rdx
2945 assert_different_registers(index, recv, rax, rdx);
2946
2947 // Test for an invoke of a final method
2948 Label notFinal;
2949 __ movl(rax, flags);
2950 __ andl(rax, (1 << ConstantPoolCacheEntry::vfinalMethod));
2951 __ jcc(Assembler::zero, notFinal);
2952
2953 Register method = index; // method must be rbx,
2954 assert(method == rbx, "methodOop must be rbx, for interpreter calling convention");
2955
2956 // do the call - the index is actually the method to call
2957 __ verify_oop(method);
2958
2959 // It's final, need a null check here!
2960 __ null_check(recv);
2961
2962 // profile this call
2963 __ profile_final_call(rax);
2964
2965 __ jump_from_interpreted(method, rax);
2966
2967 __ bind(notFinal);
2968
2969 // get receiver klass
2970 __ null_check(recv, oopDesc::klass_offset_in_bytes());
2971 // Keep recv in rcx for callee expects it there
2972 __ movptr(rax, Address(recv, oopDesc::klass_offset_in_bytes()));
2973 __ verify_oop(rax);
2974
2975 // profile this call
2976 __ profile_virtual_call(rax, rdi, rdx);
2977
2978 // get target methodOop & entry point
2979 const int base = instanceKlass::vtable_start_offset() * wordSize;
2980 assert(vtableEntry::size() * wordSize == 4, "adjust the scaling in the code below");
2981 __ movptr(method, Address(rax, index, Address::times_ptr, base + vtableEntry::method_offset_in_bytes()));
2982 __ jump_from_interpreted(method, rdx);
2983 }
2984
2985
2986 void TemplateTable::invokevirtual(int byte_no) {
2987 transition(vtos, vtos);
2988 assert(byte_no == f2_byte, "use this argument");
2989 prepare_invoke(rbx, noreg, byte_no);
2990
2991 // rbx,: index
2992 // rcx: receiver
2993 // rdx: flags
2994
2995 invokevirtual_helper(rbx, rcx, rdx);
2996 }
2997
2998
2999 void TemplateTable::invokespecial(int byte_no) {
3000 transition(vtos, vtos);
3001 assert(byte_no == f1_byte, "use this argument");
3002 prepare_invoke(rbx, noreg, byte_no);
3003 // do the call
3004 __ verify_oop(rbx);
3005 __ profile_call(rax);
3006 __ jump_from_interpreted(rbx, rax);
3007 }
3008
3009
3010 void TemplateTable::invokestatic(int byte_no) {
3011 transition(vtos, vtos);
3012 assert(byte_no == f1_byte, "use this argument");
3013 prepare_invoke(rbx, noreg, byte_no);
3014 // do the call
3015 __ verify_oop(rbx);
3016 __ profile_call(rax);
3017 __ jump_from_interpreted(rbx, rax);
3018 }
3019
3020
3021 void TemplateTable::fast_invokevfinal(int byte_no) {
3022 transition(vtos, vtos);
3023 assert(byte_no == f2_byte, "use this argument");
3024 __ stop("fast_invokevfinal not used on x86");
3025 }
3026
3027
3028 void TemplateTable::invokeinterface(int byte_no) {
3029 transition(vtos, vtos);
3030 assert(byte_no == f1_byte, "use this argument");
3031 prepare_invoke(rax, rbx, byte_no);
3032
3033 // rax,: Interface
3034 // rbx,: index
3035 // rcx: receiver
3036 // rdx: flags
3037
3038 // Special case of invokeinterface called for virtual method of
3039 // java.lang.Object. See cpCacheOop.cpp for details.
3040 // This code isn't produced by javac, but could be produced by
3041 // another compliant java compiler.
3042 Label notMethod;
3043 __ movl(rdi, rdx);
3044 __ andl(rdi, (1 << ConstantPoolCacheEntry::methodInterface));
3045 __ jcc(Assembler::zero, notMethod);
3046
3047 invokevirtual_helper(rbx, rcx, rdx);
3048 __ bind(notMethod);
3049
3050 // Get receiver klass into rdx - also a null check
3051 __ restore_locals(); // restore rdi
3052 __ movptr(rdx, Address(rcx, oopDesc::klass_offset_in_bytes()));
3053 __ verify_oop(rdx);
3054
3055 // profile this call
3056 __ profile_virtual_call(rdx, rsi, rdi);
3057
3058 Label no_such_interface, no_such_method;
3059
3060 __ lookup_interface_method(// inputs: rec. class, interface, itable index
3061 rdx, rax, rbx,
3062 // outputs: method, scan temp. reg
3063 rbx, rsi,
3064 no_such_interface);
3065
3066 // rbx,: methodOop to call
3067 // rcx: receiver
3068 // Check for abstract method error
3069 // Note: This should be done more efficiently via a throw_abstract_method_error
3070 // interpreter entry point and a conditional jump to it in case of a null
3071 // method.
3072 __ testptr(rbx, rbx);
3073 __ jcc(Assembler::zero, no_such_method);
3074
3075 // do the call
3076 // rcx: receiver
3077 // rbx,: methodOop
3078 __ jump_from_interpreted(rbx, rdx);
3079 __ should_not_reach_here();
3080
3081 // exception handling code follows...
3082 // note: must restore interpreter registers to canonical
3083 // state for exception handling to work correctly!
3084
3085 __ bind(no_such_method);
3086 // throw exception
3087 __ pop(rbx); // pop return address (pushed by prepare_invoke)
3088 __ restore_bcp(); // rsi must be correct for exception handler (was destroyed)
3089 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
3090 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError));
3091 // the call_VM checks for exception, so we should never return here.
3092 __ should_not_reach_here();
3093
3094 __ bind(no_such_interface);
3095 // throw exception
3096 __ pop(rbx); // pop return address (pushed by prepare_invoke)
3097 __ restore_bcp(); // rsi must be correct for exception handler (was destroyed)
3098 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
3099 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
3100 InterpreterRuntime::throw_IncompatibleClassChangeError));
3101 // the call_VM checks for exception, so we should never return here.
3102 __ should_not_reach_here();
3103 }
3104
3105 void TemplateTable::invokedynamic(int byte_no) {
3106 transition(vtos, vtos);
3107
3108 if (!EnableInvokeDynamic) {
3109 // We should not encounter this bytecode if !EnableInvokeDynamic.
3110 // The verifier will stop it. However, if we get past the verifier,
3111 // this will stop the thread in a reasonable way, without crashing the JVM.
3112 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
3113 InterpreterRuntime::throw_IncompatibleClassChangeError));
3114 // the call_VM checks for exception, so we should never return here.
3115 __ should_not_reach_here();
3116 return;
3117 }
3118
3119 assert(byte_no == f1_oop, "use this argument");
3120 prepare_invoke(rax, rbx, byte_no);
3121
3122 // rax: CallSite object (f1)
3123 // rbx: unused (f2)
3124 // rcx: receiver address
3125 // rdx: flags (unused)
3126
3127 Register rax_callsite = rax;
3128 Register rcx_method_handle = rcx;
3129
3130 if (ProfileInterpreter) {
3131 // %%% should make a type profile for any invokedynamic that takes a ref argument
3132 // profile this call
3133 __ profile_call(rsi);
3134 }
3135
3136 __ movptr(rcx_method_handle, Address(rax_callsite, __ delayed_value(java_dyn_CallSite::target_offset_in_bytes, rcx)));
3137 __ null_check(rcx_method_handle);
3138 __ prepare_to_jump_from_interpreted();
3139 __ jump_to_method_handle_entry(rcx_method_handle, rdx);
3140 }
3141
3142 //----------------------------------------------------------------------------------------------------
3143 // Allocation
3144
3145 void TemplateTable::_new() {
3146 transition(vtos, atos);
3147 __ get_unsigned_2_byte_index_at_bcp(rdx, 1);
3148 Label slow_case;
3149 Label slow_case_no_pop;
3150 Label done;
3151 Label initialize_header;
3152 Label initialize_object; // including clearing the fields
3153 Label allocate_shared;
3154
3155 __ get_cpool_and_tags(rcx, rax);
3156
3157 // Make sure the class we're about to instantiate has been resolved.
3158 // This is done before loading instanceKlass to be consistent with the order
3159 // how Constant Pool is updated (see constantPoolOopDesc::klass_at_put)
3160 const int tags_offset = typeArrayOopDesc::header_size(T_BYTE) * wordSize;
3161 __ cmpb(Address(rax, rdx, Address::times_1, tags_offset), JVM_CONSTANT_Class);
3162 __ jcc(Assembler::notEqual, slow_case_no_pop);
3163
3164 // get instanceKlass
3165 __ movptr(rcx, Address(rcx, rdx, Address::times_ptr, sizeof(constantPoolOopDesc)));
3166 __ push(rcx); // save the contexts of klass for initializing the header
3167
3168 // make sure klass is initialized & doesn't have finalizer
3169 // make sure klass is fully initialized
3170 __ cmpl(Address(rcx, instanceKlass::init_state_offset_in_bytes() + sizeof(oopDesc)), instanceKlass::fully_initialized);
3171 __ jcc(Assembler::notEqual, slow_case);
3172
3173 // get instance_size in instanceKlass (scaled to a count of bytes)
3174 __ movl(rdx, Address(rcx, Klass::layout_helper_offset_in_bytes() + sizeof(oopDesc)));
3175 // test to see if it has a finalizer or is malformed in some way
3176 __ testl(rdx, Klass::_lh_instance_slow_path_bit);
3177 __ jcc(Assembler::notZero, slow_case);
3178
3179 //
3180 // Allocate the instance
3181 // 1) Try to allocate in the TLAB
3182 // 2) if fail and the object is large allocate in the shared Eden
3183 // 3) if the above fails (or is not applicable), go to a slow case
3184 // (creates a new TLAB, etc.)
3185
3186 const bool allow_shared_alloc =
3187 Universe::heap()->supports_inline_contig_alloc() && !CMSIncrementalMode;
3188
3189 if (UseTLAB) {
3190 const Register thread = rcx;
3191
3192 __ get_thread(thread);
3193 __ movptr(rax, Address(thread, in_bytes(JavaThread::tlab_top_offset())));
3194 __ lea(rbx, Address(rax, rdx, Address::times_1));
3195 __ cmpptr(rbx, Address(thread, in_bytes(JavaThread::tlab_end_offset())));
3196 __ jcc(Assembler::above, allow_shared_alloc ? allocate_shared : slow_case);
3197 __ movptr(Address(thread, in_bytes(JavaThread::tlab_top_offset())), rbx);
3198 if (ZeroTLAB) {
3199 // the fields have been already cleared
3200 __ jmp(initialize_header);
3201 } else {
3202 // initialize both the header and fields
3203 __ jmp(initialize_object);
3204 }
3205 }
3206
3207 // Allocation in the shared Eden, if allowed.
3208 //
3209 // rdx: instance size in bytes
3210 if (allow_shared_alloc) {
3211 __ bind(allocate_shared);
3212
3213 ExternalAddress heap_top((address)Universe::heap()->top_addr());
3214
3215 Label retry;
3216 __ bind(retry);
3217 __ movptr(rax, heap_top);
3218 __ lea(rbx, Address(rax, rdx, Address::times_1));
3219 __ cmpptr(rbx, ExternalAddress((address)Universe::heap()->end_addr()));
3220 __ jcc(Assembler::above, slow_case);
3221
3222 // Compare rax, with the top addr, and if still equal, store the new
3223 // top addr in rbx, at the address of the top addr pointer. Sets ZF if was
3224 // equal, and clears it otherwise. Use lock prefix for atomicity on MPs.
3225 //
3226 // rax,: object begin
3227 // rbx,: object end
3228 // rdx: instance size in bytes
3229 __ locked_cmpxchgptr(rbx, heap_top);
3230
3231 // if someone beat us on the allocation, try again, otherwise continue
3232 __ jcc(Assembler::notEqual, retry);
3233 }
3234
3235 if (UseTLAB || Universe::heap()->supports_inline_contig_alloc()) {
3236 // The object is initialized before the header. If the object size is
3237 // zero, go directly to the header initialization.
3238 __ bind(initialize_object);
3239 __ decrement(rdx, sizeof(oopDesc));
3240 __ jcc(Assembler::zero, initialize_header);
3241
3242 // Initialize topmost object field, divide rdx by 8, check if odd and
3243 // test if zero.
3244 __ xorl(rcx, rcx); // use zero reg to clear memory (shorter code)
3245 __ shrl(rdx, LogBytesPerLong); // divide by 2*oopSize and set carry flag if odd
3246
3247 // rdx must have been multiple of 8
3248 #ifdef ASSERT
3249 // make sure rdx was multiple of 8
3250 Label L;
3251 // Ignore partial flag stall after shrl() since it is debug VM
3252 __ jccb(Assembler::carryClear, L);
3253 __ stop("object size is not multiple of 2 - adjust this code");
3254 __ bind(L);
3255 // rdx must be > 0, no extra check needed here
3256 #endif
3257
3258 // initialize remaining object fields: rdx was a multiple of 8
3259 { Label loop;
3260 __ bind(loop);
3261 __ movptr(Address(rax, rdx, Address::times_8, sizeof(oopDesc) - 1*oopSize), rcx);
3262 NOT_LP64(__ movptr(Address(rax, rdx, Address::times_8, sizeof(oopDesc) - 2*oopSize), rcx));
3263 __ decrement(rdx);
3264 __ jcc(Assembler::notZero, loop);
3265 }
3266
3267 // initialize object header only.
3268 __ bind(initialize_header);
3269 if (UseBiasedLocking) {
3270 __ pop(rcx); // get saved klass back in the register.
3271 __ movptr(rbx, Address(rcx, Klass::prototype_header_offset_in_bytes() + klassOopDesc::klass_part_offset_in_bytes()));
3272 __ movptr(Address(rax, oopDesc::mark_offset_in_bytes ()), rbx);
3273 } else {
3274 __ movptr(Address(rax, oopDesc::mark_offset_in_bytes ()),
3275 (int32_t)markOopDesc::prototype()); // header
3276 __ pop(rcx); // get saved klass back in the register.
3277 }
3278 __ movptr(Address(rax, oopDesc::klass_offset_in_bytes()), rcx); // klass
3279
3280 {
3281 SkipIfEqual skip_if(_masm, &DTraceAllocProbes, 0);
3282 // Trigger dtrace event for fastpath
3283 __ push(atos);
3284 __ call_VM_leaf(
3285 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc), rax);
3286 __ pop(atos);
3287 }
3288
3289 __ jmp(done);
3290 }
3291
3292 // slow case
3293 __ bind(slow_case);
3294 __ pop(rcx); // restore stack pointer to what it was when we came in.
3295 __ bind(slow_case_no_pop);
3296 __ get_constant_pool(rax);
3297 __ get_unsigned_2_byte_index_at_bcp(rdx, 1);
3298 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), rax, rdx);
3299
3300 // continue
3301 __ bind(done);
3302 }
3303
3304
3305 void TemplateTable::newarray() {
3306 transition(itos, atos);
3307 __ push_i(rax); // make sure everything is on the stack
3308 __ load_unsigned_byte(rdx, at_bcp(1));
3309 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray), rdx, rax);
3310 __ pop_i(rdx); // discard size
3311 }
3312
3313
3314 void TemplateTable::anewarray() {
3315 transition(itos, atos);
3316 __ get_unsigned_2_byte_index_at_bcp(rdx, 1);
3317 __ get_constant_pool(rcx);
3318 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray), rcx, rdx, rax);
3319 }
3320
3321
3322 void TemplateTable::arraylength() {
3323 transition(atos, itos);
3324 __ null_check(rax, arrayOopDesc::length_offset_in_bytes());
3325 __ movl(rax, Address(rax, arrayOopDesc::length_offset_in_bytes()));
3326 }
3327
3328
3329 void TemplateTable::checkcast() {
3330 transition(atos, atos);
3331 Label done, is_null, ok_is_subtype, quicked, resolved;
3332 __ testptr(rax, rax); // Object is in EAX
3333 __ jcc(Assembler::zero, is_null);
3334
3335 // Get cpool & tags index
3336 __ get_cpool_and_tags(rcx, rdx); // ECX=cpool, EDX=tags array
3337 __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // EBX=index
3338 // See if bytecode has already been quicked
3339 __ cmpb(Address(rdx, rbx, Address::times_1, typeArrayOopDesc::header_size(T_BYTE) * wordSize), JVM_CONSTANT_Class);
3340 __ jcc(Assembler::equal, quicked);
3341
3342 __ push(atos);
3343 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc) );
3344 __ pop_ptr(rdx);
3345 __ jmpb(resolved);
3346
3347 // Get superklass in EAX and subklass in EBX
3348 __ bind(quicked);
3349 __ mov(rdx, rax); // Save object in EDX; EAX needed for subtype check
3350 __ movptr(rax, Address(rcx, rbx, Address::times_ptr, sizeof(constantPoolOopDesc)));
3351
3352 __ bind(resolved);
3353 __ movptr(rbx, Address(rdx, oopDesc::klass_offset_in_bytes()));
3354
3355 // Generate subtype check. Blows ECX. Resets EDI. Object in EDX.
3356 // Superklass in EAX. Subklass in EBX.
3357 __ gen_subtype_check( rbx, ok_is_subtype );
3358
3359 // Come here on failure
3360 __ push(rdx);
3361 // object is at TOS
3362 __ jump(ExternalAddress(Interpreter::_throw_ClassCastException_entry));
3363
3364 // Come here on success
3365 __ bind(ok_is_subtype);
3366 __ mov(rax,rdx); // Restore object in EDX
3367
3368 // Collect counts on whether this check-cast sees NULLs a lot or not.
3369 if (ProfileInterpreter) {
3370 __ jmp(done);
3371 __ bind(is_null);
3372 __ profile_null_seen(rcx);
3373 } else {
3374 __ bind(is_null); // same as 'done'
3375 }
3376 __ bind(done);
3377 }
3378
3379
3380 void TemplateTable::instanceof() {
3381 transition(atos, itos);
3382 Label done, is_null, ok_is_subtype, quicked, resolved;
3383 __ testptr(rax, rax);
3384 __ jcc(Assembler::zero, is_null);
3385
3386 // Get cpool & tags index
3387 __ get_cpool_and_tags(rcx, rdx); // ECX=cpool, EDX=tags array
3388 __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // EBX=index
3389 // See if bytecode has already been quicked
3390 __ cmpb(Address(rdx, rbx, Address::times_1, typeArrayOopDesc::header_size(T_BYTE) * wordSize), JVM_CONSTANT_Class);
3391 __ jcc(Assembler::equal, quicked);
3392
3393 __ push(atos);
3394 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc) );
3395 __ pop_ptr(rdx);
3396 __ movptr(rdx, Address(rdx, oopDesc::klass_offset_in_bytes()));
3397 __ jmp(resolved);
3398
3399 // Get superklass in EAX and subklass in EDX
3400 __ bind(quicked);
3401 __ movptr(rdx, Address(rax, oopDesc::klass_offset_in_bytes()));
3402 __ movptr(rax, Address(rcx, rbx, Address::times_ptr, sizeof(constantPoolOopDesc)));
3403
3404 __ bind(resolved);
3405
3406 // Generate subtype check. Blows ECX. Resets EDI.
3407 // Superklass in EAX. Subklass in EDX.
3408 __ gen_subtype_check( rdx, ok_is_subtype );
3409
3410 // Come here on failure
3411 __ xorl(rax,rax);
3412 __ jmpb(done);
3413 // Come here on success
3414 __ bind(ok_is_subtype);
3415 __ movl(rax, 1);
3416
3417 // Collect counts on whether this test sees NULLs a lot or not.
3418 if (ProfileInterpreter) {
3419 __ jmp(done);
3420 __ bind(is_null);
3421 __ profile_null_seen(rcx);
3422 } else {
3423 __ bind(is_null); // same as 'done'
3424 }
3425 __ bind(done);
3426 // rax, = 0: obj == NULL or obj is not an instanceof the specified klass
3427 // rax, = 1: obj != NULL and obj is an instanceof the specified klass
3428 }
3429
3430
3431 //----------------------------------------------------------------------------------------------------
3432 // Breakpoints
3433 void TemplateTable::_breakpoint() {
3434
3435 // Note: We get here even if we are single stepping..
3436 // jbug inists on setting breakpoints at every bytecode
3437 // even if we are in single step mode.
3438
3439 transition(vtos, vtos);
3440
3441 // get the unpatched byte code
3442 __ get_method(rcx);
3443 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::get_original_bytecode_at), rcx, rsi);
3444 __ mov(rbx, rax);
3445
3446 // post the breakpoint event
3447 __ get_method(rcx);
3448 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint), rcx, rsi);
3449
3450 // complete the execution of original bytecode
3451 __ dispatch_only_normal(vtos);
3452 }
3453
3454
3455 //----------------------------------------------------------------------------------------------------
3456 // Exceptions
3457
3458 void TemplateTable::athrow() {
3459 transition(atos, vtos);
3460 __ null_check(rax);
3461 __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
3462 }
3463
3464
3465 //----------------------------------------------------------------------------------------------------
3466 // Synchronization
3467 //
3468 // Note: monitorenter & exit are symmetric routines; which is reflected
3469 // in the assembly code structure as well
3470 //
3471 // Stack layout:
3472 //
3473 // [expressions ] <--- rsp = expression stack top
3474 // ..
3475 // [expressions ]
3476 // [monitor entry] <--- monitor block top = expression stack bot
3477 // ..
3478 // [monitor entry]
3479 // [frame data ] <--- monitor block bot
3480 // ...
3481 // [saved rbp, ] <--- rbp,
3482
3483
3484 void TemplateTable::monitorenter() {
3485 transition(atos, vtos);
3486
3487 // check for NULL object
3488 __ null_check(rax);
3489
3490 const Address monitor_block_top(rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
3491 const Address monitor_block_bot(rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
3492 const int entry_size = ( frame::interpreter_frame_monitor_size() * wordSize);
3493 Label allocated;
3494
3495 // initialize entry pointer
3496 __ xorl(rdx, rdx); // points to free slot or NULL
3497
3498 // find a free slot in the monitor block (result in rdx)
3499 { Label entry, loop, exit;
3500 __ movptr(rcx, monitor_block_top); // points to current entry, starting with top-most entry
3501 __ lea(rbx, monitor_block_bot); // points to word before bottom of monitor block
3502 __ jmpb(entry);
3503
3504 __ bind(loop);
3505 __ cmpptr(Address(rcx, BasicObjectLock::obj_offset_in_bytes()), (int32_t)NULL_WORD); // check if current entry is used
3506
3507 // TODO - need new func here - kbt
3508 if (VM_Version::supports_cmov()) {
3509 __ cmov(Assembler::equal, rdx, rcx); // if not used then remember entry in rdx
3510 } else {
3511 Label L;
3512 __ jccb(Assembler::notEqual, L);
3513 __ mov(rdx, rcx); // if not used then remember entry in rdx
3514 __ bind(L);
3515 }
3516 __ cmpptr(rax, Address(rcx, BasicObjectLock::obj_offset_in_bytes())); // check if current entry is for same object
3517 __ jccb(Assembler::equal, exit); // if same object then stop searching
3518 __ addptr(rcx, entry_size); // otherwise advance to next entry
3519 __ bind(entry);
3520 __ cmpptr(rcx, rbx); // check if bottom reached
3521 __ jcc(Assembler::notEqual, loop); // if not at bottom then check this entry
3522 __ bind(exit);
3523 }
3524
3525 __ testptr(rdx, rdx); // check if a slot has been found
3526 __ jccb(Assembler::notZero, allocated); // if found, continue with that one
3527
3528 // allocate one if there's no free slot
3529 { Label entry, loop;
3530 // 1. compute new pointers // rsp: old expression stack top
3531 __ movptr(rdx, monitor_block_bot); // rdx: old expression stack bottom
3532 __ subptr(rsp, entry_size); // move expression stack top
3533 __ subptr(rdx, entry_size); // move expression stack bottom
3534 __ mov(rcx, rsp); // set start value for copy loop
3535 __ movptr(monitor_block_bot, rdx); // set new monitor block top
3536 __ jmp(entry);
3537 // 2. move expression stack contents
3538 __ bind(loop);
3539 __ movptr(rbx, Address(rcx, entry_size)); // load expression stack word from old location
3540 __ movptr(Address(rcx, 0), rbx); // and store it at new location
3541 __ addptr(rcx, wordSize); // advance to next word
3542 __ bind(entry);
3543 __ cmpptr(rcx, rdx); // check if bottom reached
3544 __ jcc(Assembler::notEqual, loop); // if not at bottom then copy next word
3545 }
3546
3547 // call run-time routine
3548 // rdx: points to monitor entry
3549 __ bind(allocated);
3550
3551 // Increment bcp to point to the next bytecode, so exception handling for async. exceptions work correctly.
3552 // The object has already been poped from the stack, so the expression stack looks correct.
3553 __ increment(rsi);
3554
3555 __ movptr(Address(rdx, BasicObjectLock::obj_offset_in_bytes()), rax); // store object
3556 __ lock_object(rdx);
3557
3558 // check to make sure this monitor doesn't cause stack overflow after locking
3559 __ save_bcp(); // in case of exception
3560 __ generate_stack_overflow_check(0);
3561
3562 // The bcp has already been incremented. Just need to dispatch to next instruction.
3563 __ dispatch_next(vtos);
3564 }
3565
3566
3567 void TemplateTable::monitorexit() {
3568 transition(atos, vtos);
3569
3570 // check for NULL object
3571 __ null_check(rax);
3572
3573 const Address monitor_block_top(rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
3574 const Address monitor_block_bot(rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
3575 const int entry_size = ( frame::interpreter_frame_monitor_size() * wordSize);
3576 Label found;
3577
3578 // find matching slot
3579 { Label entry, loop;
3580 __ movptr(rdx, monitor_block_top); // points to current entry, starting with top-most entry
3581 __ lea(rbx, monitor_block_bot); // points to word before bottom of monitor block
3582 __ jmpb(entry);
3583
3584 __ bind(loop);
3585 __ cmpptr(rax, Address(rdx, BasicObjectLock::obj_offset_in_bytes())); // check if current entry is for same object
3586 __ jcc(Assembler::equal, found); // if same object then stop searching
3587 __ addptr(rdx, entry_size); // otherwise advance to next entry
3588 __ bind(entry);
3589 __ cmpptr(rdx, rbx); // check if bottom reached
3590 __ jcc(Assembler::notEqual, loop); // if not at bottom then check this entry
3591 }
3592
3593 // error handling. Unlocking was not block-structured
3594 Label end;
3595 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
3596 __ should_not_reach_here();
3597
3598 // call run-time routine
3599 // rcx: points to monitor entry
3600 __ bind(found);
3601 __ push_ptr(rax); // make sure object is on stack (contract with oopMaps)
3602 __ unlock_object(rdx);
3603 __ pop_ptr(rax); // discard object
3604 __ bind(end);
3605 }
3606
3607
3608 //----------------------------------------------------------------------------------------------------
3609 // Wide instructions
3610
3611 void TemplateTable::wide() {
3612 transition(vtos, vtos);
3613 __ load_unsigned_byte(rbx, at_bcp(1));
3614 ExternalAddress wtable((address)Interpreter::_wentry_point);
3615 __ jump(ArrayAddress(wtable, Address(noreg, rbx, Address::times_ptr)));
3616 // Note: the rsi increment step is part of the individual wide bytecode implementations
3617 }
3618
3619
3620 //----------------------------------------------------------------------------------------------------
3621 // Multi arrays
3622
3623 void TemplateTable::multianewarray() {
3624 transition(vtos, atos);
3625 __ load_unsigned_byte(rax, at_bcp(3)); // get number of dimensions
3626 // last dim is on top of stack; we want address of first one:
3627 // first_addr = last_addr + (ndims - 1) * stackElementSize - 1*wordsize
3628 // the latter wordSize to point to the beginning of the array.
3629 __ lea( rax, Address(rsp, rax, Interpreter::stackElementScale(), -wordSize));
3630 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray), rax); // pass in rax,
3631 __ load_unsigned_byte(rbx, at_bcp(3));
3632 __ lea(rsp, Address(rsp, rbx, Interpreter::stackElementScale())); // get rid of counts
3633 }
3634
3635 #endif /* !CC_INTERP */